Portable Diagnostics for Visual Function
2 likes536 views
The document discusses the future of healthcare through the lens of artificial intelligence and machine learning, highlighting its impact on diagnostics, patient care, drug development, and system design. It outlines the barriers and risks associated with implementing new technologies in the healthcare sector, emphasizing the importance of data quality and the ethical implications of advancements in medicine. Additionally, it reflects on the need for healthcare professionals to adapt to these changes and the evolving roles of algorithms in clinical decision-making.
![Eye
Movements
Useful in virtual
reality environments
in general
Allows foveated rendering
(LoD, level-of-detail),
which again reduces
computation requirements
for convincing VR
rendering
Foveated rendering is a process that combines eye tracking
and software to adjust the way a VR experience is rendered in
realtime. Withfoveatedrendering, thePC runningyourVivewith
7invensun eye tracker only has to render the greatest detail in
thesmallareaon whichyoureyesaredirectlyfocused.
https://uploadvr.com/7invensun-eye-tracker-for-vive/
Games like this don’t just look incredible because of ‘hyper-realism’ but because
their engineers use all sorts of tricks [LOD’ing, or Level of Detail; Mipmapping;
frustumculling,etc.] to savememory.
https://kotaku.com/horizon-zero-dawn-uses-all-sorts-of-clever-tricks-to-lo-1
794385026](https://image.slidesharecdn.com/portablevisualfunctiondiagnostics-180409071608/85/Portable-Diagnostics-for-Visual-Function-47-320.jpg)
![Magneto-
retinography
With Diamond
Magnetometry
(instead of costly SQUID
magnetometers used in
magnetoencephalography
[MEG] for example).
At some point eventually allowing
non-contact electrical
measurements when the price goes
even well below $150k per
instrument?
Today, Matthew Dale and Gavin
Morley at the University of
Warwick in the U.K. say that
diamond sensors are poised to
revolutionize the way physicians
use magnetic field measurements
in diagnostic medicine. They map
out the state of the art in this area
and say that the business
opportunityissignificant.
https://arxiv.org/abs/1705.01994
https://www.technologyreview.com/s/607871/how-diamond-sensors-are-set-to-revolutionize-medical-diagnostics/
There are around 100 SQUID MEG systems installed worldwide, at a cost of over $1M each. The MCG
market should be much larger if the instrumentation was affordable and portable, because MCG has been
shown to be superior to ECG and hence other non-invasive approaches for the diagnosis of coronary
artery disease (CAD) [Kwongetal.2013, Fenici et al. 2005 and 2013]. CAD is the most common type of heart
diseaseandistheleadingcauseofdeathintheUnitedStatesinbothmenandwomen.
Several companies have tried and failed to commercialize SQUID-based MCG, held back by the cost
of a cryogen-based system. We estimate that 100,000 MCG systems could be sold if the functionality
were the same as existing SQUID systems and the price was below $150k. This is based on there being
over 100,000hospitalsinChina,India,theEU,JapanandtheUSA.
Diamond magnetometers are at technology readiness level (TRL) 7: the technology has been
demonstrated and is moving towards being put on sale. However, this has not yet reached the sensitivity
neededfor MCG,soanMCGsystembasedondiamondisatTRL4-5(technologydevelopment).](https://image.slidesharecdn.com/portablevisualfunctiondiagnostics-180409071608/85/Portable-Diagnostics-for-Visual-Function-55-320.jpg)
![Intraocular
Pressure
Boucard et al. 2016: “The classic view of
glaucomab
is that of an eye disease in
which elevated intraocular pressure (IOP)
mechanically damages the optic nerve (ON)
causing the death of retinal ganglion cells
(RGCs). Indeed, in high-pressure glaucoma
(HPG, the most common form of
glaucoma), RGC and ON damage are
associated with an elevated IOP
(>21 mmHg).[1]
However, this view cannot
be complete as glaucoma with normal
levels of IOP is commonly reported as well.
In such normal-pressure glaucoma (NPG),
damage occurs to the ON without the eye
pressure exceeding the normal range. By
definition, NPG only differs from HPG in
that the IOP is consistently below 22
mmHG.[1]
Moreover, rather than being a
disease restricted to the eye, damage of
the RCGs extends to the axons that form
the primary visual pathways.c
The patient wears the SENSIMED Triggerfish® system up to 24 hours
and assumes normal activities including sleep periods. The SENSIMED
Triggerfish® Sensor is a soft disposable silicone contact lens
embedding a micro-sensor that captures spontaneous circumferential
changes at the corneoscleral area.
http://www.sensimed.ch/
Journal ofGlaucoma. August 22, 2016.
doi: 10.1097/IJG.0000000000000517
The range of IOP fluctuation was larger in the eyes with
normal-tension glaucoma (NTG) than in the nonglaucoma
eyes. This larger fluctuation might be one of the reasons
underlying the aggravation of the visual field by NTG.
Measurements of 24-hour continuous IOP might be one of
the useful methods to distinguish NTG from nonglaucoma
eyes.
Daily variation of intraocular pressure
Measurement once in 6 months might
not capture all relevant information
IOP Transient
Cyclical strain vs. constant strain
(i.e. punching your eye with fist vs. applying
constant pressure over longer time)
IOP variability by CrawfordDowns atWorld Glaucoma Conference2017
HelsinkiFinland, at“Newfrontiersinglaucoma”session,Saturday July1, 2017
https://youtu.be/1nrV3zztisk |https://youtu.be/QNDzq5Rp5RA
IOP vs. CSFO
Intraocular pressure, cerebrospinal pressure
or trans lamina pressure difference or what?
http://dx.doi.org/10.1016/j.preteyeres.2015.01.002](https://image.slidesharecdn.com/portablevisualfunctiondiagnostics-180409071608/85/Portable-Diagnostics-for-Visual-Function-61-320.jpg)
Portable Diagnostics for Visual Function
- 1. Petteri Teikari, PhD http://petteri-teikari.com/ Version “Mon 9 April 2018 “ Portable Visual Function Diagnostics Deep learning based data- driven ophthalmology beyond unimodal “magical” scalar measures
- 2. Future Trends for Healthcare at HealthtechFundingForum-Advancing Innovationin DigitalHealth September 27, 2017|WellcomeTrust, London, UK DrVishalGulatiResponsibleforhealthcaredealsatDraperEspritplc, alistedPatientCapitalVC firm: interpretedfor“HealthcareDesign”: “If one would design healthcare systems now, they would not look like the contemporary ones. … just look at how Chinese are automating their healthcare.. and how startups that target health in general rather than solving problems at the hospital only when people have got already sick” Lost in Thought — The Limitsof the Human Mind and the Future of Medicine Ziad Obermeyer, M.D., and ThomasH. Lee,M.D. NEngl J Med 2017; 377:1209-1211 September28, 2017 DOI: 10.1056/NEJMp1705348 If a root cause of our challenges is complexity, the solutions are unlikely to be simple. Asking doctors to work harder or get smarter won’t help. There is little doubt that algorithms will transform the thinking underlying medicine. The only question is whether this transformation will be driven by forces from within or outside the field. If medicine wishes to stay in control of its own future, physicians will not only have to embrace algorithms, they will also have to excel at developing and evaluating them, bringing machine-learning methods into the medical domain.
- 3. Future Trends for Healthcare AIIsYourDoctor’sNextBest Friend MikeMcCormickJan192017 https://mccormick.vc/ai-is-your-doctors-next-best-friend-2bb33e7cf4e8 Wherewillintelligentmachines affect healthcare? Eventuallymachine intelligencewill touch virtuallyall aspectsof healthcare. Four areasalready beingaffected are… ● Diagnosticsand detection: Examples:radiology, tissueanalysis, genomicinsights, chatbots, and patient monitoringviaexternal sensors, wearables and implantables. ● Treatment and patient care: Examples: personalized precision drugsand treatment plans, remotepatient monitoring, and automated real-time treatmentadjustments. ● Drug development: Deep learningwill augment the pharmaceutical industry’sincreasinglycostlyR&D processesby identifyingpatternsin molecular interactionsat previouslyunheard oflevels of granularityand efficiency. Machine learningwill also better match patientstoclinical trialsleadingtobetter patient outcomesand faster drugapprovals. ● Informatics, system-design and data management: Machineswill bringefficiencytothe interactionsthat takeplace within the complexweb of stakeholdersand processesthatcomprise modern healthcare systems. Barriersandrisk factors Healthcare, perhapsmorethan mostindustries, presentsseveral barriers and risk factorsto newtechnologiesand would-bedisruptors: Highstakes: Theliteral life-and-deathnatureof healthcaremakefor a tinymarginof error inpatient-facing technologies. Legal andregulatory issues: Healthcareisamong theworld’smost heavilyregulatedindustries. Datasecurity andaccess: Making data accessibleyet secureis crucial. Dataquality: Thoughthehealthcareindustryissitting on ever-growing mountainsof data, thequalityand relevanceof thedatasets isn’talways great,and accessing meaningful datasetscan bechallenging,particularly forstartups. Causalcomplexity inbiology anddisease:Ourunderstanding of biological systemsand diseasesisincomplete. Thecellularprogression of cancersand thecomplexityof moment-to-momentneural interactions,forexample, areprocesseswe’refarfromfully understanding. Misalignedincentives: Disparatestakeholdersarenotalways incentivizedtosharedataorplaynicelywithoneanother. Bioethicalconsiderations:Theseissueswillbecomestickier and moredifficultto parseasthelinesbetween biologyand technologyblur. Forexample, whatwill betheethical implicationsof advanced genetic engineering thatallowsfor“designer”babiescrafted to theirparents’ exactspecifications?
- 4. Future Trends for Healthcare Slowly FDA (and regulators in general) is waking up to the situation "When you start adding analytical AI for any image analysis—think of detecting cancer or some other serious disease—at that point people need to know when that detection means something and is real," Bakul Patel, FDA’s associate director for digital health, says. https://spectrum.ieee.org/the-human-os/biomedi cal/devices/fda-assembles-team-to-oversee-ai-r evolution-in-health New AI Device for Diabetes Eye Screening to Complete FDA ClinicalTrial IDx, an early-stage medical device company focused on developing software-based algorithms that can identify disease in medical images is currently conducting an FDA clinical trial to obtain clearance for its first product, IDx-DR by the end of summer 2017. IDx-DR is a screening solution for diabetic retinopathy. IDx also has algorithms in development for the detection of macular degeneration, glaucoma, Alzheimer’s disease, cardiovascular disease, and stroke risk. http://hitconsultant.net/2017/07/06/new-ai-device-diabetes-eye-screening/
- 5. Future Trends for Healthcare Digitalizing hospital processes The Hospitalof the Future isa Network February17,2017– JeroenTas ChiefInnovation&StrategyOfficer atPhilips https://www.linkedin.com/pulse/hospital-future-network- connecting-care-continuous-health-jeroen-tas/ HowGoogleDeepMind'sStreamsapp islayingthefoundationsforartificial intelligence-powered healthcarein theNHS Wednesday 10May2017 http://www.cityam.com/264463/google-deepminds-streams-app-laying-foundation s-artificial DeepMind's work with the NHS to help alert doctors to patients whose health is at risk is laying the groundwork for delivering information that one day will be powered by artificial intelligence. The Google-owned British pioneer is working with the Royal Free London NHS Trust on a smartphone app called Streams. It currently uses an NHS created algorithm to provide information to clinicians relating to acute kidney injury (AKI), with DeepMind creating the method of delivery that includes "breaking news" style notifications. It does not use AI despite the companies expertise in the technology. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5333321/ Gettingtheinteroperabilityanddatapipe solved allowing 3rd party“AIapps”andreally allowingefficientdeeplearning miningofpatients Through FHIR (“Fire API”)
- 6. Future Trends for Healthcare System Thinking rather than module- based optimization reducing system efficiency https://hbr.org/2017/06/hospitals-are-dramatically-overpaying- for-their-technology -revolutionise-way-they-work-how-hospitals-could-be-rebuilt-better CommandCenter to ImprovePatientFlow http://www.hopkinsmedicine.org/news/articles/command-center-to-improve-patient-flow https://www.ahcmedia.com/articles/139933-hopkins-command-center-improves-quality-with-coordination http://www.modernhealthcare.com/article/20161126/MAGAZINE/311269980
- 7. Future Trends for Healthcare Deep learning image analysis for just one type images, let alone full EHR mining is constrained by poor data infrastructure and bad curation with missing segmentation and pathology class labels. Labeling need medical expertise making the process harder than just crowdsourcing dog vs. cat labels for example CrowdsourcingtoEvaluateFundusPhotographsforthe PresenceofGlaucoma Wang et al. (2017) doi: 10.1097/IJG.0000000000000660 To assess the accuracy of crowdsourcing for grading optic nerve images for glaucoma using Amazon Mechanical Turk before and after training modules. Gamificationof theelectron microscopesegmentationthrough EyeWireprojectrun by Sebastian Seung. A VideogameThatRecruitsPlayersto Map theBrain |WIRED EyeWire,A Gameto Map theBrain fromMIT Voxeleron Orion has developed a good augmented intelligence for efficient collaboration with the AI and the person segmenting the retinal layers (without yet gamifying theexperience) By ROWLAND MANTHORPE 23 Sep 2017 http://www.wired.co.uk/article/harri-valpola-curious- ai-artificial-intelligence-third-wave Harri Valpola, 44, is founder of TheCuriousAI Company (co-founded with Antti Rasmus, Timo Haanpää and Mathias Berglund), that focuses on semi- supervised learning a 20-person artificial intelligence startup based in Helsinki, which has just raised $3.67 million in funding – small change compared to many tech funding rounds, but an impressive sum for a company that has no products and is onlyinterested in research. https://doi.org/10.1016/j.patcog.2016.09.030
- 8. Future Trends for Healthcare Rethinking medicine – novel ways to deliver healthcare by ChristinaFarr April 7, 2017 https://www.technologyreview.com/s/604053/can-digital-therapeutics-be-as-good-as-drugs/ Jose Hamilton: “The "real" digital therapy won't be a competitor to biological therapy. But a platform where psychological treatments (digital or personal), biological (pills) or even physical would be optimized, personalized and accountable.“ To distinguishthemselvesfrom“wellness”gadgets, digitaltherapeuticscompaniestendto carryout clinicaltestsandsometimesseekregulatory approvals
- 9. Future Trends for Healthcare Rethinking medicine – how clinical profession should change, with the ones not changing, perishing away Digital evangelists argue that intelligent machines will be able to incorporate the latest data and research immediately, but that is both questionable and a potential weakness. Clinical trials vary in scale and quality, and indiscriminate inclusion would inevitably lead to mistakes. Digital hardliners would argue that machines should judge the quality of the research, but for the foreseeable future the expertise of doctors will be essential to deciding the validity of new approaches. So perhaps one of the most powerful effects of artificial intelligence will be, perversely, to make healthcare more human and personal. It will remove the dependency on doctors’ fallible memory and incomplete knowledge, and free them to use machine-generated information to work with patients to shape their specific treatment. 11 March 2017 https://www.theguardian.com/healthcare-network/2017/mar/11/artificial-intelligence-nhs-doctor-patient-relationship https://www.newyorker.com/magazine/2017/04/03/ai-versus-md GeoffreyHinton now qualifies the provocation. “The role of radiologists will evolve from doing perceptual things that could probably be done by a highly trained pigeon to doing far more cognitive things,” he told me. His prognosis for the future of automated medicine is based on a simple principle: “Take any old classification problem where you have a lot of data, and it’s going to be solved by deep learning. There’s going to be thousands ofapplicationsofdeeplearning.”
- 10. Future Trends for Healthcare Rethinking medicine – making doctors more human again https://www.technologyreview.com/s/609060/put-humans-at-the-center-of-ai/
- 11. Future Trends for Healthcare All the “digital natives” will get into the play January11,2017 Nokia'svisionfordigitalhealth:FromAIanalyticstoconnectedhairbrushes.NokiaboughtFrenchhealthdevice manufacturerWithingsearlierthisyeartotakeon the IoThealthcaremarket.What'snext forthecompany. http://www.zdnet.com/article/nokias-vision-for-digital-health-from-ai-analytics-to-connected-hairbrushes/ January24,2017: The Chan Zuckerberg Inititative,aphilanthropic initiativefrom Facebook CEOMark ZuckerbergandhiswifeDr.Priscilla Chan,apediatrician,hasacquiredastartup, Meta,focusedon usingAIandmachinelearningto sift throughrecently published scientificstudies. TheChan Zuckerberg Initiativeisalimitedliabilitycompanyfocusedon the ambitious goal to"cure,prevent,ormanagealldiseasesby theendof thecentury." Atleast$3billionwillbeallocatedtowardthat goal,allcomingoutofChan andZuckerberg'sFacebookshares. http://www.mobihealthnews.com/content/chan-zuckerberg-initiative-acquires-ai-startup-meta-will-offer-its-services-free April30,2017: Googletocommercializeartificial intelligencetodetect diseases.LilyPeng,productmanagerofthemedical imagingteamatGoogleResearch,sharedhowtheUStechgiantisusing deeplearningtotrain machinesto analyze medicalimagesand automaticallydetectpathologicalcues,beitswollen bloodvesselsintheeye orcancerous tumors,duringavideoconferencewiththeSouthKorean mediahostedby GoogleKorea. http://m.theinvestor.co.kr/view.php?ud=20170430000162 June7,2017: Applewantsa pieceof theartificialintelligencepie. Apple’sResearchKit,whichusesiPhonesto collect health information and then makesthedataavailableforresearch, isshowingpromiseafterscientistspublisheddataon seizures,asthmaattacksandheartdiseaseusingthetool.WhileApplestillfaceschallengesapplyingResearchKit’s resultstoabroaderpopulation (mostconsumersofAppleproductsareyounger,well-off and well-educated),the company seemsdetermined tocarveoutanichein healthcareandAIcould helpitsefforts. http://www.healthcaredive.com/news/apple-wants-a-piece-of-the-artificial-intelligence-pie/444393/ July28, 2017: Here'swhattomakeofAmazon'spotentialconnected healthplay …Amazon'spotentialadvantagesinthe connectedhealthdevicemarketlikelyoutnumberit'sdisadvantages. http://uk.businessinsider.com/amazons-healthcare-play-2017-7?r=US&IR=T September25, 2017: Microsoft hires Iain Buchan , worldleaderindigitalhealthcaretotakepersonalisedhealthtothenextlevel https://news.microsoft.com/en-gb/2017/09/25/microsoft-hires-world-leader-in-digital-healthcare-to-take-personalised-health-to-the-next-level/
- 12. Future Trends for Healthcare Data is the new gold And for the healthcare (especially public systems), to the stay competitive, the organizations should think of their data strategy along with monetization schemes https://hbr.org/2017/06/to-survive-health-care-data-providers-need-to-stop-selling-data “Most data-driven healthcare IT (HCIT) providers aren’t going to survive. Their business models are at serious risk of failure in the next three to five years. To beat those odds, they need to evolve dramatically, and fast, to a point where they are not selling data at all.”
- 13. Future Trends for Eye Care
- 14. Future Trends for Eye Care Similar slow awakening to “machine medicine” in ophthalmology as in healthcare in general BIGDATA:CURRENTSTATUSANDFUTURE DIRECTIONSAGENDA ARVO2017|Baltimore,MD Organizers:MichaelF.Chiang,MD,AnneL.Coleman,MD,PhD,FARVOandSethBlackshaw,PhD “Similarly as highlighted in the perspective by Obermeyer and Lee (2017) in previous slide, ophthalmologist training need to keep up-to-date with the machine learning revolution. With big data, the role of diagnoses have to be re-valued as well as even more fine-grained phenotyping becomes possible. And would a patient with DR and glaucoma be just DR+Glaucoma, or something slightly different with these coexisting pathologies”
- 15. Future Trends for Eye Care AI-driven drug discovery for eye care as well SiliconValleyComputational DrugStartupTakeson Glaucoma By TeklaS.Perry-Posted 13Mar 2017 https://spectrum.ieee.org/view-from-the-valley/at-work/start-ups/silicon-valley-computational-drug-startup-takes-on-glaucoma TwoXAR (AndrewA.Radin) announced a partnership with Santen Inc., the U.S. subsidiary of Japanese ophthalmology company Santen Pharmaceutical, to collaborate on identifying new drug candidates for the treatmentofglaucoma. BenevolentAI is currently largest privateAI firm in Europe 21March2017 https://www.cnbc.com/video/2017/03/21/be nevolent-ai-is-currently-largest-private -ai-firm-in-europe.html http://www.wired.co.uk/article/benevolent-ai-london-unicorn-pharma-startup
- 16. Future Trends for Eye Care Battle of the Egos on many fronts. Optometrists want to to upskill themselves, and essentially make money from surgeries and lucrative VEGF injections “Both optometrists and ophthalmologists point to experiences in Oklahoma to support their positions. Bryant said that in Oklahoma, which has allowed expanded work for optometrists the longest, there were only two reported complaints for more than 25,000 procedures. The ophthalmologists point to a research paper published last October in the medical journal JAMA Ophthalmology (Steinetal.2016) that found that patients who had a certain type of laser surgery to treat glaucoma had to go back for treatment on the same eye 35.9 percent of the time when an optometrist did the work, as opposed to 15.1 percent of the time when an ophthalmologist did it.” http://www.newsobserver.com/news/politics-government/state-polit ics/article131198204.html MAR02,2017-AAO OptometristsinFloridaTake BrazenStepTowardPrimary- CareProviderStatus Proposal sets the bar for audacious assaults on patient safety by attempting to place 100,000-plus non-surgeons on equal footing withophthalmologists http://optometrytimes.modernmedicine.com/optometrytimes/news/intravitreal- injections-optometrists The article, “Implementation of aNurse-Delivered Intravitreal Injection Service” was published in the June 2014 issue of Eye. The purpose of this study was “to introduce nurse-delivered intravitreal injections to increase medical retina treatment capacity in the United Kingdom.” … “Our preliminary results of a series of 4,000 nurse-delivered injections associated without serious vision-threatening complication is indicative that this procedure can be safely administered by a nurse.” No cases of post-intravitreal anti-VEGFendophthalmitisoccurred in thisstudy.
- 17. Future Trends for Eye Care Maintaining healthy lifestyle as the most obvious first step to treat. Not every approach needs to be high-tech and highly scalable digital service AerobicExerciseforNeuroprotection "Aerobic exerciseis known to lower intraocular pressure(IOP), which weknow protects retinal ganglion cells," says Harry A. Quigley, MD, professorand director of glaucoma services at the WilmerEye Institute at Johns Hopkins University in Baltimore. "And short-term studies show it may improve blood flow to theretina and optic nerveas well." http://www.glaucoma.org/treatment/aerobic-exercise-for-neuroprotection.php http://dx.doi.org/10.1111/acel.12512 “These data provide new insight into the mechanisms underlying exercise-mediated protection of retinal cells. We found that daily forced exercise, initiated 24 h after an acute RGC-specific injury in middle-aged mice, led to a substantial improvement in RGC function and survival.” Lifestyle,Nutrition,and Glaucoma LouisR Pasquale,JaeHeeKang Journal of Glaucoma: August2009 -Volume18 - Issue6 - pp 423-428 doi: 10.1097/IJG.0b013e31818d3899 In this review, we have examined the evidence on whether environmental factors are related to developing glaucoma. How do we answer the questions from newly diagnosed glaucoma patients on lifestyle behaviors and their relation to POAG? There is even scarcer data on lifestyle factors and their influence on disease progression. However, rather than default to the view that patients should simply comply with medical therapy and follow-up recommendations (which of course is true), we also suggest advocating for activities consistent with overall good health such as avoidance of smoking, moderate exercise and a diet high in fruits and vegetables. The weight of the current medical is not sufficiently strong to make broad recommendations regarding activities that glaucoma patients should avoid because they elevate IOP such as certain yoga positions, playing high wind instruments for long periods of time, and drinking large amountsof caffeinated coffee.
- 18. Future Trends for Eye Care Where is the innovation happening? According to the 2017 Centre for World University Rankings (CWUR) rankings by subject, UCL Institute of Ophthalmology (Moorfields Hospital) is thebest place in the world to study ophthalmology http://cwur.org/2017/subjects.php#Ophthalmology University College London (Moorfields) Harvard University Johns Hopkins University (Wilmer Eye Institute) University of Melbourne National University of Singapore (Singapore Eye Research Institute, SERI) University of Sydney University of Southern California University of Miami (Bascom Palmer Eye Institute) University of California, Los Angeles (UCLA Stein Eye Institute Westwood) University of California, San Diego (Shiley Eye Institute)
- 20. de facto “Standard” of portable future https://theophthalmologist.com/issues/01 16/the-eye-exams-quantum-leap/ http://dx.doi.org/10.1167/tvst.6.4.16 Test 6: Pupillometry Pupil reactions were assessed using simultaneous OCT capture of the anterior segments including the iris plane. Each eye was stimulated independently and sequentially with a single, bright, 250-ms flash of white light. B- scan recordings are captured at regular intervals of 350 ms prior to stimulation and 4000 ms post-stimulation. Measurements of the pupil circumference could subsequently be calculated to identify pupil abnormalities and relative afferentpupillarydefects
- 21. User Experience Portable Diagnostics Easier to carry around just one single device doingmost ofthe stuff rather than dedicated devices for each task Inhospitalsettings,onecanthenuse dedicateddevicesforhigherdiagnostic capability after the portable“pre- screening”
- 23. Fundus Imaging with some variants Annidis RHAMultispectralfundusimagingsystem http://www.annidis.com/page/technology OptomedAurora portablefundusimaging http://www.annidis.com/page/technology OptosUltra-widefundus imaging ”High-endimaging” http://www.nikon.com/about/technology/pr oduct/retinal-imaging/index.htm Do it yourself smartphone fundus camera – DIYretCAM https://dx.doi.org/10.4103%2F0301-4 738.194325 (a) The do it yourself smartphone fundus camera used as a hand held the device. (b) The do it yourself smartphone fundus camera can be held at the condensing lens and supported with the other hand on the camera. (c) Like in indirect ophthalmoscopy, scleral depression is done after stabilizing the do it yourself smartphone fundus camera Multi-spectralimagingfor invivoimaging of oxygentensionand -amyloidβ Dr. TosTJM Berendschot, Prof.dr. Carroll ABWebers UniversityEyeClinicMaastricht
- 24. Fundus Imaging Smartphones getting more and more ubiquitous. Clip-ons and embedded electronics allow cloud-based teleophthalmology both with automated AI approaches and augmented with human expert ophthalmologists for hard-to- reach areas Peek Visiondeveloped a smartphoneapp and lensattachment, Peek Retina,to capturesharp imagesof theback oftheeye. https://www.fastcompany.com/3062154/smartphones-are-leading-the-global-charge-against-blindness Nonmydriatic Fundus Camera Based on the Raspberry Pi® Computer” paper in J Ophth. He says the camera can be replicated from parts easily foundonline for about $185. An ophthalmology resident at the University of Illinois at Chicago College of Medicine has invented an inexpensive, handheld camera that can photograph the retina without need for pupil dilation. https://blogs.nvidia.com/blog/2016/02/17/deep-learning-4/ . https://youtu.be/wa9OdaRMgO8 As the founder of SocialEyes, Nicholas Bedworth is delivering healthcare via embedded NVIDIA (such as the JetsonTX2) at scale in places where doctors are scarce, and without internet for cloud connection EyeSpy:SocialEyesUsesDeepLearning toSpotSeriousEyeProblems Centralized cloud GPU inference allows higher throughputs when good enough internet is available for example using the V100 GPUs from NVIDIA (and the even faster generations after that)
- 25. SLO Scanning Light Ophthalmoscope Adaptive optics scanning light ophthalmoscopy (AOSLO) is an emerging technology for improving in vivo imaging of the human retinal microvasculature, allowing unprecedented visualization of retinal microvascular structure, measurements of blood flow velocity, and microvascular network mapping. Human retinal microvascular imaging using adaptive optics scanning light ophthalmoscopy Chui et al. (2016) https://doi.org/10.1186/s40942-016-0037-8 Distribution differences of macular cones measured by AOSLO: Variation in slope from fovea to periphery more pronounced than differences in total cones Ann E. Elsner, Toco Y.P. Chui, Lei Feng, Hong Xin Song, Joel A. Papay, Stephen A. Burns (2017) https://doi.org/10.1016/j.visres.2016.06.015 Cone density variesamongindividualsbymore than just ascalar factor. Imaging Foveal Microvasculature: Optical Coherence Tomography Angiography Versus Adaptive Optics Scanning Light Ophthalmoscope Fluorescein Angiography Shelley Mo; Brian Krawitz; Eleni Efstathiadis; Lawrence Geyman; Rishard Weitz; Toco Y. P. Chui; Joseph Carroll; Alfredo Dubra; Richard B. Rosen (2016) http://dx.doi.org/10.1167/iovs.15-18932 Optical coherence tomography angiography is comparable to AOSLO FA at imaging the foveal microvasculature except for differences in FAZ area, lumen diameter, and some qualitative features. These results, together with its ease of use, short acquisition time, and avoidance of potentially phototoxic blue light, support OCTA as a tool for monitoring ocular pathology and detecting early disease. Photoreceptor-Based Biomarkers in AOSLO Retinal Imaging Katie M. Litts; Robert F. Cooper; Jacque L. Duncan; Joseph Carroll (2017) http://dx.doi.org/10.1167/iovs.17-21868 Resolving cone inner and outer segment structure with AOSLO. Shown are confocal (A) and split- detection (B) images from the parafoveal retina of a patient with CNGA3-associated ACHM. The color-merged image (C) has the confocal image displayed in green and the split-detection image in red.
- 26. OCT with various variants again (2016) https://doi.org/10.1364/BOE.7.001783 https://dx.doi.org/10.1167/tvst.3.3.10 Biomedical OpticsExpressVol. 8, Issue 4, pp. 2287-2300 (2017) https://doi.org/10.1364/BOE.8.002287
- 27. Intelligent Imaging Embed deep learning to the device and the image acquisition process to minimize operator- dependent image quality degradations (see e.g. OSCAR-IB study) Think of the consumer AI-driven camera systems as inspiration Better for the patient and the operator that the camera automatically re-acquires the images and even reconstructs the image from partially good quality shots rather than realizing the suboptimal quality later when patient have left the hospital already 4October 2017: Google has announced a new add- on for the Pixel 2 camera called Clips. The camera is hands-free and works like a photographer you’ve hired for an event. The AI captures moments for you, and then you decide how to use the images later when you can look through them. An AI engine snaps photos when you arenotevenlooking or paying attention. https://www.fastcompany.com/3059281/introducing-hover-an-ai-powered-indoor -safe-camera-drone Plan3D:Viewpointand TrajectoryOptimization for AerialMulti-View StereoReconstruction BenjaminHepp, MatthiasNießner, Otmar Hilliges (Submittedon25May2017) https://arxiv.org/abs/1705.09314
- 28. Imaging In the wild – in “high street” optometry and in hospitals OCTROLLOUTINEVERY SPECSAVERS ANNOUNCED Themultiplewillensureall740ofitsUK practiceshaveanOCTdevice installedwithin thenexttwoyears.22May2017 by Emily McCormick https://www.aop.org.uk/ot/industry/high-street/2017/05/22/oct -rollout-in-every-specsavers-announced WHAT'SGOINGONIN TECH? Revealingthelatesttechnologylaunchesforthe practice07 Mar2017 by LaurenceDerbyshire https://www.aop.org.uk/ot/industry/equipment-and-suppliers/20 17/03/07/whats-going-on-in-tech https://youtu.be/JS550fDKyOE OT spoke to a range of equipment companies (Optos, Heidelberg, Carl Zeiss and Bib Ophthalmic Instruments) and experts at 100% Optical to find out about the latest technologylaunches forthe practice. In practice opticians do not have enough skilled staff for image analysis, and the whole imaging process should be made deep learning -driven and high quality always independent of the device operator Thiscreatesopportunitiestosell automatedimageanalysissoftwarefor opticians, reducing their laborcosts, improving diagnosticquality andcreating cross-selling opportunities. https://www.mi vision.com.au/ taking-financi al-control-of- your-practice/
- 29. Imaging Beyond Retinopathies Retina as being the most easily measured part of the brain, makes sense to use it as a proxy for other pathologies EyeScans toDetect CancerandAlzheimer’s Disease TheHumanOS |Biomedical|Diagnostics ByMeganScudellariPosted31Aug2017 https://spectrum.ieee.org/the-human-os/biomedical/diagnostics/eye-sca ns-to-detect-cancer-and-alzheimers-disease At the University of Washington, a team led by computer scientist ShwetakPatel created a smartphone app (BiliScreen) to screen for pancreatic cancer with a quick selfie. Developed over the last year and ahalf, the team recently tested their system in a clinicalstudyof70people. They were able to identify cases of concern with 89.7 percentsensitivityand96.8percentaccuracy. At Cedars-Sinai and NeuroVision Imaging LLC in California, researchers have developed a sophisticated camera and retinal imaging approach to detect early signs of Alzheimer’s disease (AD). The camera capture beta-amyloid plaques being taggedwith curcumin. This system, recentlydetailed in a proof-of-concepttrial published in the journal JCI Insight, relieson aspecialized ophthalmic camerathatisnotyetavailableonasmartphone. Photo:Dennis Wise/Universityof Washington
- 31. Ocular Blood Flow with Laser speckle flowgraphy (LSFG) or AOSLO https://doi.org/10.1155/2017/2969064 ChangesinRetinal Vessel Architecture andBlood Flow in Multiple Sclerosis(P6.401) Richard Nicholas, Adam Dubis, Ashwini Nandoskar, Jeremy Chataway and John Greenwood http://www.neurology.org/content/88/16_Supplement/P6.401.short WaveformAnalysisof OcularBlood FlowandtheEarlyDetectionof NormalTensionGlaucoma Shiga et al. (2013) IOVS doi:10.1167/iovs.13-12930 Softcare Co.,Ltd.ispleased to announcethat it received 510(k) clearancefrom the U.S. Foodand DrugAdministration (FDA)for LSFG-NAVI. http://www.softcare-ltd.co.jp/510k_clearance.html BostonMicromachines ApaerosRetinalImagingSystem – Small-vessel blood flow http://www.bostonmicromachines.com/qualitative-measures-of-small-vessel-blood-flow-.html
- 32. Pupillometry Technically“simple” method thatcan be deployedeasily into avirtualreality headset or into a smartphoneclip-on Headsetkeepsthe eye-cameradistance constant (without themoreexpensivetelecentric lenses thatis) ,and reducing the error ofthe samepupil size toappear smaller/largertodistance variations.
- 33. Pupillometry Evolution of the technique Chopraetal.(2017) http://dx.doi.org/10.1167/tvst.6.4.16 OCT-based pupillometry Murray etal.(1981) https://doi.org/10.1016/0165-0270(81)90024-8 Video-based pupillometry Starket al.(1959) https://doi.org/10.1109/JRPROC.1959.287206 Iris reflectivity-based pupillometry 850-950 nm infrared lighting for example from LEDs is used illuminate the eye, and the pupil boundary is computed from the video via computer vision techniques
- 34. Pupillometry Components Temporalresolution High-freqencybiomarkersrequire highsampling ates Spatialresolution Low-amplitudebiomarkersrequire morepixelsfromcamera Blinkandartifactsingeneral “Realpupilsize” Measured pupilsize -Discretization noise (in thiscase, pupil width in pixels) -Instrumentation noise (image noise, and uncertainty ofboundaryalgorithm) -Physiological noise (“focusfluctuations”, and mental processes) -“Pupilnoise” (“pupillaryunrest” used asaindex of alertnessfor example, Stanten et al. 1966)
- 35. Pupillometry Spatial and temporal characteristics of pupillary size Pupil noise (hippus, pupillary unrest) is characterized as a random noise on top of the “mean pupil size”-signal in the frequency range of 0.05-0.3 Hz (Stark 1959; Usui andStark 1982). Experiments have indicated skewing of the pupil noise spectrum from Gaussian white noise at high and low pupil areas. This skewing has been attributed the multiplier gain dependence on the expansive range nonlinearity and length-tension relationship of iris muscles (Usui and Stark 1978; Usui andStark1982) . In practice, pupil neuromuscular dynamics shape the high frequency cutoff for noise pulses and large and small sinusoids, whereas retinal adaptation accounts for the low- frequency asymptotes(Stanten andStark 1966;Stark 1984). Usui andStark 1982 Stark1959
- 36. Pupillometry Spatial and temporal characteristics of pupillary size Adhikari et al. (2016) http://dx.doi.org/10.1038/srep33373 Maynard et al. (2015) http://dx.doi.org/10.1167/iovs.15-17357 Pupil size do not change especially fast, and the modulation frequencies (e.g. sinusoidal light stimuli) are not very high especially for melanopsin studies. Thus, it makes more sense to have slower sampling rates and higher resolution cameras if onecannot get theboth for reasonable price.
- 37. Pupillometry Example of “spatial” PLR use, also beyond retinopathies Alzheimer’sdiseaseinthehumaneye.Clinicalteststhat identifyocularand visualinformationprocessingdeficitasbiomarkers ChangL.Y.L.,LoweJ.,ArdilesA.,LimJ.,GreyA.C.,Robertson K.,etal..(2014)..AlzheimersDement.10,251–261. 10.1016/j.jalz.2013.06.004 Methodfollowupofabove:InfraredVideoPupillography CoupledwithSmartPhone LEDforMeasurementofPupillary LightReflex LilyYu-LiChang, Jason Turuwhenua, Tian Yuan Qu, JoannaM. Black, and MonicaL. Acosta Front Integr Neurosci. 2017;11:6.Published online 2017 Mar 7. doi: 10.3389/fnint.2017.00006 with Point GreyFireflyUSB 2.0 infrared camera(752 x480, 60fps, $275) CorrelationsBetweenHourlyPupillometerReadingsandIntracranial Pressure Values McNett,Molly;Moran,Cristina; Janki,Clare;Gianakis,Anastasia,JournalofNeuroscienceNursing:August2017- Volume49-Issue4 -p229–234.10.1097/JNN.0000000000000290 Theuse anduptake of pupillometersinthe Intensive Care Unit MatthewHaoLee,BiswadevMitra,JiunKaePui,MarkFitzgerald AustralianCriticalCareAvailableonline17July2017 https://doi.org/10.1016/j.aucc.2017.06.003 Useof DigitalPupillometrytoMeasure SedativeResponsetoPropofol Haddocketal(2017)TheOchsner Journal:Fall2017,Vol.17,No.3,pp.250-253 http://www.ochsnerjournal.org/doi/abs/10.1043/1524-5012-17.3.250?code=occl-site Infrared pupillometryhelpstodetectandpredictdeliriumin thepost- anesthesiacare unit EricYang,MatthiasKreuzer,SeptemberHesse, ParanDavari,SimonC.Lee,PaulS.García JournalofClinicalMonitoringandComputing(2017), https://doi.org/10.1007/s10877-017-0009-z
- 38. Pupillometry Focus on non-invasive trauma evaluation Highlighting also the increased use of crowdfunding platform for scientific uses. Waiting for the “Scientific ICO” rollouts still... Dr.CharleneOng,atHarvardUniversityHospitalsMassachusetts General and Brigham and Women’s, hopes to find a safe and effectivemethodtocatchsignsofbrainswellingearlier.
- 39. Pupillometry Example of “temporal” PLR use, also beyond retinopathies Pupillary Motility: Bringing Neuroscience to the Psychiatry Clinicof the Future SimonaGraurandGregSiegle CurrentNeurologyand NeuroscienceReports August2013, 13:365 https://doi.org/10.1007/s11910-013-0365-0 Reduced Pupillary Unrest: Autonomic NervousSystemAbnormality inDiabetesMellitus AstradurBHreidarsson andHansJorgen GGundersenDiabetes 1988Apr; 37(4):446-451 https://doi.org/10.2337/diab.37.4.446 Fatigue and cognition: Pupillary responses to problem solving inearly multiple sclerosis‐ patients REgg, B Högl, SGlatzl, RBeer, TBerger MultipleSclerosisJournalVolume:8 issue: 3,page(s):256-260 https://doi.org/10.1191/1352458502ms793oa Using pupilsize and heartrate to inferaffective statesduring behavioral neurophysiology and neuropsychology experiments SigridA.deRodezBenaventetal.| Brain andBehavior 2017 http://dx.doi.org/10.1002/brb3.717 Pupillary unrest correlateswith arousal symptoms and motorsignsin Parkinsondisease SamayJain etal.(2011)MovementDisorders http://dx.doi.org/10.1002/mds.23628 Comparisonofthe antidepressantsreboxetine, fluvoxamine and amitriptyline upon spontaneous pupillary fluctuationsin healthy humanvolunteers M.A.Phillips,P.Bitsios,E.Szabadi,C.M.Bradshaw Psychopharmacology March2000, Volume149, Issue 1, pp72–76 https://doi.org/10.1007/s002139900334 Assessing Pain Using the VariationCoefficient ofPupillary Diameter DavidJ. Charieretal.(2017)TheJournalof Pain Availableonline13 July2017 https://doi.org/10.1016/j.jpain.2017.06.006
- 40. Pupillometry Camera selection in practice 33 series-GigE monochromeindustrial cameras https://www.theimagingsource.com/products/industrial-cameras/gige-monochrome/ -Windows and Linux softwareincluded ~$730 ~$330 ~$439 ~$495 ~$730 oemcameras.com by MicahSingleton @MicahSingleton Feb 7,2017, 11:15amEST https://www.theverge.com/circuitbreaker/2017/2/7/14532610/sony-smartphone-camera-sensor-1000-fps https://www.sony.net/SonyInfo/News/Press/201702/17-013E/index.html Ultra-HighResolutionMonochrome Cameras http://www.adept.net.au/cameras/ultraMono.shtml Genie NANO XL-M51005120x 5120 px, CMOS Mono, 20fps, GigE HS-20000M 5120 x3840 px, CMOS, 32fps,10GigE Flare 48MP30-CX 7920x 6004px, CMOS, 30.9 fps EoSens25CXP+ 5120x 5120 px,CMOS, 80 fps, CXP-6 CoaXPress CP80-25-M-72 5120 x5120px, CMOS,72fps, CoaXPress PhantomVEO4K PL forfilm 4096 x2304px,CMOS, 1000 fps IOIndustriesFlare48MPCameraDemo https://youtu.be/WvW9532k81M https://youtu.be/dFdU-JjypWs
- 42. Pupillometry Camera Calibration and comparison rig If you want to become a PLR Powerhouse lab, you could measure the same subjects in clinical settings for example with three different quality levels and learn quality improvement with deep learning? 1) Very low-cost cameraphone 2a) Entry-levelindustrialcamera 2b) High-end smartphone 3) High-end industrialcamera See similar idea for training adeep learning network for3D reconstruction from indoor scans Slide 10 of https://www.slideshare.net/PetteriTeikari PhD/dataset-creation-for-deep-learningbas ed-geometric-computer-vision-problems An exampleof “path multiplication” with a beamsplitter fora slit lamp setup. This allows simultaneous exam by the clinician andrecording of theexam
- 43. Pupillometry Commercial landscape Established:NeurOptics NPi®-200Pupillometer athttps://www.neurocriticalcare.org/ Open-source: PupilLabs eyetracker withHoloLens,HTCViveandOculusadd-onsavailable -https://doi.org/10.1145/2638728.2641695 Emerging: “ThePUPIL Study”-Automated, QuantitativePupil Assessment Using BinocularOCT Sponsor: UniversityCollege, London (PearseKeane) https://clinicaltrials.gov/ct2/ show/NCT03081468 Emerging: “PupilScreen: Using Smartphonesto AssessTraumaticBrain Injury. Mariakisetal. (2017) http://doi.org/10.1145/3 131896 Emerging: ”BrightLamp use thecamera and torch of your phoneto measureyour pupil's responseto lightstimulusto diagnoseconcussion.” www.brightlamp.org techcrunch.com/2017/05/16 “Wehaveatotalof8 peopleontheteam:2 computervision specialists,regulatory, financial,legal,app dev,marketing, ML/diagnostics.Allof whichhavevery specificrolesin growingastartup.” reddit.com
- 44. Eye Movements Similar “end-to-end” pipeline as for pupillometry (and even the same hardware) One can “easily” integrate various measures to a single VR headset https://github.com/pupil-labs/hmd-eyes Orloskyetal.(2017) https://doi.org/10.1109/TVCG.2017.2657018
- 45. Eye Movements Even Oculus joins the Open Source development kit Theopensourcereleaseofthe RiftDevelopmentKit2(DK2). hardwarefollowsonfromourearlier releasesof RiftDK1 and LatencyTester.Thisincludes schematics,boardlayout, mechanicalCAD,artwork,and specificationsunder aCreative CommonsAttribution4.0license, andfirmwareunderBSD+PATENT licences.Wepresentaguidedtour ofDK2for thoseinterestedin digging indeeper. https://developer.oculus.com/blog/open-source-release-of-rift-dk2/
- 47. Eye Movements Useful in virtual reality environments in general Allows foveated rendering (LoD, level-of-detail), which again reduces computation requirements for convincing VR rendering Foveated rendering is a process that combines eye tracking and software to adjust the way a VR experience is rendered in realtime. Withfoveatedrendering, thePC runningyourVivewith 7invensun eye tracker only has to render the greatest detail in thesmallareaon whichyoureyesaredirectlyfocused. https://uploadvr.com/7invensun-eye-tracker-for-vive/ Games like this don’t just look incredible because of ‘hyper-realism’ but because their engineers use all sorts of tricks [LOD’ing, or Level of Detail; Mipmapping; frustumculling,etc.] to savememory. https://kotaku.com/horizon-zero-dawn-uses-all-sorts-of-clever-tricks-to-lo-1 794385026
- 48. Eye Movements Beyond retinopathies http://doi.org/10.1002/mds.27105 Many high-prevalence neurological disorders involve dysfunctions of oculomotor control and attention, including attention deficit hyperactivity disorder (ADHD), fetal alcohol spectrum disorder (FASD), and Parkinson’sdisease(PD). https://doi.org/10.1007/s00415-012-6631-2 https://doi.org/10.1016/j.bpsc.2016.12.009 https://doi.org/10.1016/j.cobeha.2016.03.008 https://doi.org/10.1016/j.bandc.2008.08.026
- 49. Eye Movements Commercial landscape Unlockingthe potential of eye trackingtechnology Posted Feb 19,2017 by BenDickson (@bendee983) https://techcrunch.com/2017/02/19/unlocking-the-potential-of-eye-tracking-technology/ A newbrainhealthappfrom Neurotrackwarns users of memory decline Posted Dec1,2016 by LoraKolodny (@lorakolodny) https://techcrunch.com/2016/12/01/neurotrack-takes-brain-scans-home/ https://qz.com/604397/a-simple-five-minute-test-could-make-earlier-diagno sis-of-alzheimers-possible/ Oculusacquireseye-tracking startup TheEye Tribe 28Dec2016- Thestartuphasdevelopeda$99 eye tracking devicedeveloper kits https://techcrunch.com/2016/12/28/the-eye-tribe-oculus/ The iPhone 8 Could Soon Usher inEye-TrackingAds Sep5th,2017 https://www.macobserver.com/columns-opinions /editorial/iphone-8-eye-tracking-ads/ Better drugs could also lead to “better quality of life for the patient, which would ultimately lower health care costs,” says Samir Kaul, the founding managing director of Khosla Ventures. Khosla is the lead investor for Neurotrack’s $6.5 million in new funding, announced Jan. 27. If it passes muster in ongoing studies, Neurotrack’s online test could be a cheap, easy, non-invasive way to detect Alzheimer’s in advance of symptoms. The Palo Alto, California-based company is inviting physicians to offer the eye-tracking test to their patients. It’s also developing a personalized lifestyle program for prospective patients. The program will be based on emerging research in Finland (FINGER) and elsewhere suggesting that diet, exercise, cognitive training, sleep and stress management could preserve brain health and help prevent Alzheimer’s.
- 50. Electro- physiology Electroretinography (ERG) and electro- oculogram (EOG) The systems used to be hard to use with long dark adaptations and long setup times. Recently new devices such as EvokeDx and RETEval have made things easier in practice http://konanmedical.com/evokedx/ https://youtu.be/aT6dCD_5p5k Multifocal ERG http://vsri.ucdavis.edu/research/electrophysiology Wearable electrooculography (EOG) goggles. The Swiss Federal Institute of Technology (ETH) Zurich developed these goggles to track relative eye movements. https://doi.org/10.1109/MPRV.2010.86
- 51. Electro- physiology Correlating function with structure Documenta Ophthalmologica April 2017, Volume134, Issue 2, pp 111–128| Comparingthree differentmodesof electroretinographyinexperimental glaucoma: diagnosticperformance and correlationtostructure Laura Wilsey,Sowjanya Gowrisankaran,Grant Cull, Christy Hardin.Claude F. Burgoyne,Brad Fortune https://doi.org/10.1007/s10633-017-9578-x ArqBrasOftalmol. 2017Mar-Apr;80(2):118-121. doi:10.5935/0004-2749.20170028 Structure-functionalcorrelation usingadaptiveoptics,OCT,and microperimetryinacaseofoccult macular dystrophy. VianaKÍ, MessiasA, SiqueiraRC, RodriguesMW, Jorge R Optical coherencetomography(1-3), adaptiveoptics(A-H), and microperimetry (2-4) imagesof a patientwith occultmaculardystrophy. (Arrows) Lossof continuityof theouterphotoreceptorlayerin thecentral fovealregion.(yellowasterisk)Reduced ring photoreceptor densityin the fovealregion.(2) Reductionof ringsensitivityin thecentral foveal region. (Red Asterisk)Reduced photoreceptor densityin acentralfoveal region.(4) Reduced central sensitivityinthefovea.
- 53. Electro- physiology Relatively simple to develop and integrate to the same VR-type of headset used for pupillometry and eye movement measurement A filter setting of 1–200 Hz appears most sensitive to detect glaucomatous damage if using a two-global- flash mfERG: High frequencies of 100–300 Hz also contain information that differentiates glaucoma from normal and thus should be included in the analysis. A 50 Hz notch filter allows grossly contaminated waveforms to be analyzed in a meaningful manner. With a 50 Hz filter, glaucoma patients still differed significantly from normal.
- 54. Visual Evoked Potentials Essentially a EEG headset with reduced electrode count and active dry electrodes ThenGoggle,aPortableBrain-Computer Interfacefor AssessmentofVisualFunctionandglaucoma diagnosis–Nakanishietal.(2017) http://dx.doi.org/10.1001/jamaophthalmol.2017.0738 REINVENT:Alow-cost,virtual realitybrain-computerinterface forseverestrokeupperlimb motorrecovery Spiceret al. (2017) https://doi.org/10.1109/VR.2017.7892338 AVirtual-RealityBased NeurofeedbackGame Frameworkfor Depression RehabilitationusingPervasive Three-ElectrodeEEGCollector Caietal. (2017) https://doi.org/10.1145/3127404.3127433 HTC Vive Modified With Neurable ReadsYourMind At SIGGRAPH https://youtu.be/47WHqDNckI8 https://www.extremetech.com/extreme/254816-eeg- virtual-reality-matrix-just-around-corner A feasibilitystudyonSSVEP-basedinteractionwith motivatingandimmersivevirtual andaugmented reality Josef Faller, Brendan Z. Allison, ClemensBrunner, ReinholdScherer, Dieter Schmalstieg, GertPfurtscheller, ChristaNeuper (Submitted on15Jan 2017) https://arxiv.org/abs/1701.03981
- 55. Magneto- retinography With Diamond Magnetometry (instead of costly SQUID magnetometers used in magnetoencephalography [MEG] for example). At some point eventually allowing non-contact electrical measurements when the price goes even well below $150k per instrument? Today, Matthew Dale and Gavin Morley at the University of Warwick in the U.K. say that diamond sensors are poised to revolutionize the way physicians use magnetic field measurements in diagnostic medicine. They map out the state of the art in this area and say that the business opportunityissignificant. https://arxiv.org/abs/1705.01994 https://www.technologyreview.com/s/607871/how-diamond-sensors-are-set-to-revolutionize-medical-diagnostics/ There are around 100 SQUID MEG systems installed worldwide, at a cost of over $1M each. The MCG market should be much larger if the instrumentation was affordable and portable, because MCG has been shown to be superior to ECG and hence other non-invasive approaches for the diagnosis of coronary artery disease (CAD) [Kwongetal.2013, Fenici et al. 2005 and 2013]. CAD is the most common type of heart diseaseandistheleadingcauseofdeathintheUnitedStatesinbothmenandwomen. Several companies have tried and failed to commercialize SQUID-based MCG, held back by the cost of a cryogen-based system. We estimate that 100,000 MCG systems could be sold if the functionality were the same as existing SQUID systems and the price was below $150k. This is based on there being over 100,000hospitalsinChina,India,theEU,JapanandtheUSA. Diamond magnetometers are at technology readiness level (TRL) 7: the technology has been demonstrated and is moving towards being put on sale. However, this has not yet reached the sensitivity neededfor MCG,soanMCGsystembasedondiamondisatTRL4-5(technologydevelopment).
- 56. Visual Fields Very common measurement even though it can be stressful for the patient with high noise in this psychophysical measurement (reduced by using log units) Similarly, visual field measurement fail to detect early changes in glaucoma as the brain can compensate for the neurodegeneration of RGCs ARVO 2017 poster for deep learning in visual field assessment 2846—B0449Adeep-learningbased automaticglaucomaidentification. Serife SedaS. Kucur1 , M. Abegg2 , S. Wolf2 , R. Sznitman1 . 1 ARTORGCenter, Universityof Bern, Bern, Switzerland; 2 DepartmentofOpthalmology, Inselspital Bern, Bern, Switzerland https://doi.org/10.1016/j.ophtha.2017.06.028 VisualFieldTestingwithHead- MountedPerimeter‘imo’ Matsumoto et al. (2016) https://doi.org/10.1371/journal.pone.0161974 The perimeter imo has completely isolated optical systems for the right and left eyes. Stimulus presentation is also independently performedfor eacheye. Visualfieldexaminationmethodusing virtual realityglassescomparedwith the Humphrey perimeter Tsapakis et al. (2017) doi: 10.2147/OPTH.S131160 Effect ofcognitive demand on functionalvisualfield performancein seniordriverswithglaucoma Gangeddula, Viswa, et al (2017) https://doi.org/10.3389/fnagi.2017.00286
- 57. Visual Function Towards assessing real- life function, simultaneously for diagnosis and disease progression purposes Using virtual reality to cause a subject to correct for perceived motion has revealed that glaucoma patients’ reactions are more erratic than those of healthy individuals Diniz-Filho,etal.(2015) doi: 10.1016/j.ophtha.2015.02.010 ChristopherKent,Senior EditorPUBLISHED 6JULY2015 Virtual Reality:A NewFrontierinEye Care? https://www.reviewofophthalmology.com/article/virtual-reality-a-new-frontier-in-eye-care Daga,Fábio B., et al. "Wayfinding and Glaucoma: A Virtual Reality Experiment." InvestigativeOphthalmology&Visual Science58.9(2017):3343-3349. doi: 10.1167/iovs.17-21849 The VEHuNT consisted of a cave automatic virtual environment (CAVE) used to present an immersiveVR environmentto studywayfinding tasks. Usingvirtual/augmentedrealitytosimulate visualimpairments ByDr.PeteRJones http://www.ucl.ac.uk/~smgxprj/projects.html
- 58. Vision Disorders Myopia, amblyopia, etc. Virtual reality is still far from beinga mainstream technology. But when Facebook bought VR headset-makerOculus for $2 billion last year, itsignaled to the world thatvirtual reality was no longer sci-fi, kicking off a frenzy ofexperimentation. For James Blaha, who’s struggled all his lifewith strabismus (typeofamblyopia)—a visioncondition more commonlyknown as crossedeyes—virtual realityoffereda potential cure, and he’s builta venture-backed company See Vividly (Vivid Vision software) based onthis promise. https://qz.com/489048/an-entrepreneur-is-using-virtual-reality-headsets-to-try-to-c ure-vision-disorders/ TheCure/Diagnosis and TheCause? http://dx.doi.org/10.3109/02713683.2016.1158271 https://endmyopia.org/virtual-reality-the-next-myopia-tsunami/ https://essilorusa.com/newsroom/virtual-reality-bad-fo-the-eye
- 59. Gamification of diagnosis and Service Design for Eyecare. Patients can be given the headsets for the waiting room and make their waiting time less boring and make the “process” more efficient Similarly it might be hard to get children to be attentive. The gamification might help the children to focus Amblyopia treatmentof adultswith dichoptic training usingthe virtual realityOculusRifthead mounted display:preliminary results Peter Žiak,Anders Holm, Juraj Halička, Peter Mojžišand David P Piñero BMCOphthalmology201717:105 https://doi.org/10.1186/s12886-017-0501-8
- 61. Intraocular Pressure Boucard et al. 2016: “The classic view of glaucomab is that of an eye disease in which elevated intraocular pressure (IOP) mechanically damages the optic nerve (ON) causing the death of retinal ganglion cells (RGCs). Indeed, in high-pressure glaucoma (HPG, the most common form of glaucoma), RGC and ON damage are associated with an elevated IOP (>21 mmHg).[1] However, this view cannot be complete as glaucoma with normal levels of IOP is commonly reported as well. In such normal-pressure glaucoma (NPG), damage occurs to the ON without the eye pressure exceeding the normal range. By definition, NPG only differs from HPG in that the IOP is consistently below 22 mmHG.[1] Moreover, rather than being a disease restricted to the eye, damage of the RCGs extends to the axons that form the primary visual pathways.c The patient wears the SENSIMED Triggerfish® system up to 24 hours and assumes normal activities including sleep periods. The SENSIMED Triggerfish® Sensor is a soft disposable silicone contact lens embedding a micro-sensor that captures spontaneous circumferential changes at the corneoscleral area. http://www.sensimed.ch/ Journal ofGlaucoma. August 22, 2016. doi: 10.1097/IJG.0000000000000517 The range of IOP fluctuation was larger in the eyes with normal-tension glaucoma (NTG) than in the nonglaucoma eyes. This larger fluctuation might be one of the reasons underlying the aggravation of the visual field by NTG. Measurements of 24-hour continuous IOP might be one of the useful methods to distinguish NTG from nonglaucoma eyes. Daily variation of intraocular pressure Measurement once in 6 months might not capture all relevant information IOP Transient Cyclical strain vs. constant strain (i.e. punching your eye with fist vs. applying constant pressure over longer time) IOP variability by CrawfordDowns atWorld Glaucoma Conference2017 HelsinkiFinland, at“Newfrontiersinglaucoma”session,Saturday July1, 2017 https://youtu.be/1nrV3zztisk |https://youtu.be/QNDzq5Rp5RA IOP vs. CSFO Intraocular pressure, cerebrospinal pressure or trans lamina pressure difference or what? http://dx.doi.org/10.1016/j.preteyeres.2015.01.002
- 62. Intraocular Pressure Already devices with both FDA and CE approvals for clinical use German medical device company Implandata has received CE Marking for EyeMate, the first approved IOP- monitoringsystemtooffer24-hourpressuremeasurementsinpatientswithprimary open-angleglaucoma(POAG). The sensor consists of eight pressure-sensitive capacitors and a circular microcoil antenna, which is coupled to an external handheld unit that displays readings to the patient and sends real-time data to the physician over the internet. An associated smartphoneappcan beusedto displayIOPhistoryandsetmedication alerts. https://www.aao.org/headline/continuous-iop-monitoring-implant-approved-in-euro
- 63. OMICs Genomics, proteomics, metabolomics, lipidomics, etc Progress in Retinal and EyeResearch Volume 58, May2017, Pages 89-114:15. Characterizing the “POAGome”: A bioinformatics-drivenapproach to primary open-angle glaucoma Danford etal.(2017) https://doi.org/10.1016/j.preteyeres.2017.02.001 SciRep. 2017;7:41595.https://dx.doi.org/10.1038/srep41595 AnOcular Protein Triad Can Classify FourComplex Retinal Diseases Kuiperetal. (2017) In the era where molecular assessment has improved dramatically, we aimed at the identification of biomarkers in 175 ocular fluids to classify four archetypical ocular conditions affecting the retina (age-related macular degeneration, idiopathic non- infectious uveitis, primary vitreoretinal lymphoma, and rhegmatogenous retinal detachment) with one single test. IOVS July 2017, Vol.58, BIO88-BIO98. Omics Biomarkers in Ophthalmology SusetteLauwen; Eiko K.deJong; Dirk J. Lefeber; AnnekeI. den Hollander https://doi.org/10.1167/iovs.17-21809 Here, we review the application of omics techniques in eye diseases, focusing on age- related macular degeneration (AMD), diabetic retinopathy (DR), retinal detachment (RD), myopia, glaucoma, Fuchs' corneal dystrophy (FCD), cataract, keratoconus, and dry eyes. We observe that genomic analyses were mainly successful in AMD research (almost half of the genomic heritability has been explained), whereas large parts of disease variability or risk remain unsolved in most of the other diseases. Expert Review ofOphthalmology Volume 11, 2016 - Issue 2 GWAS in myopia: insightsinto disease and implicationsforthe clinic KatieM Williams &ChristopherJ Hammond Departmentof Ophthalmology, King’s College London, London, UK;Departmentof Twin Research &Genetic Epidemiology, King’s College London, London, UK In this review we focus on what a genome-wide association study involves, what studies have been performed in relation to myopia to date, and what they ultimately tell us about myopia variance and functional pathways leading to pathogenesis. The current limitations of genome-wide association studies are reviewed and potential means to improve our understanding of the genetic factors for myopia are described. Overview of the different layers within a biological system contributing to multifactorial diseases and their relation to each other. For each layer, the name of thecorresponding omicstechniqueisindicated in blueboxes.
- 64. Unimodal Model Many proof-of-concept (PoC) published showing that deep learning can replace humans for low-level unimodal tasks such as image analysis
- 65. Unimodal Model The typical ophthalmologic approach to tackle diagnostics. Doctors are eager in finding new scalar measures that can quantify the pathology the best As much as it would be nice to have scalar variables and simple decision trees, it might not be realistic way to model complex pathogenesis Examples: OCT BMO-MRW (Kabbara et al. 2017): “To compare the cube and radial scan patterns of the spectral domain optical coherence tomography (SD-OCT) for quantifying the Bruch's membrane opening minimum rim width (BMO-MRW). The BMO-MRW diagnostic accuracy for glaucoma detectionandratesofchangederivedfromthetwoscanpatternswerecompared.” IOP (Chan et al. 2017): “A UK study of 8623 Norfolk residents has found that the use of intraocular pressure (IOP) to detect glaucoma is ‘inaccurate and probably not viable’ … The researchers also report that no single IOP threshold provided adequate sensitivity and specificity for diagnosis of glaucoma. … ‘The evidence around the performance of IOP for either screening or case- findingisnot strong,’PaulFosteremphasised.
- 66. Unimodal Diagnostics JAMA. 2016;316(22): 2402-2410. doi: 10.1001/jama.2016.17216 One million anonymised eye scans from Moorfields Eye Hospital will be used to train an artificial intelligence (AI) system from Google DeepMind. A spokesperson for DeepMind Health told Business Insider: "DeepMind has reimbursed Moorfields for the direct costs they have incurred de- personalising and manually segmenting eye scans prior to transfer … "We hope this work will help doctors faster analyse the 3,000 eye scans Moorfields carries out every week.”
- 67. Unimodal Disease Management Investigative Ophthalmology & Visual Science June 2017, Vol.58, BIO141-BIO150. doi:10.1167/iovs.17-21789
- 69. Multimodal Model Going beyond human capabilities, the next generation models will be incorporating “full clinical knowledge” of the patient
- 70. Multimodal Model Traditional numerical methods for “small data” problems might not be enough for the emergence of “network medicine” HollyF.Ainsworthetal.(2017): The use of causal inference techniques to integrate omics and GWAS data has the potential to improve biological understanding of the pathways leading to disease. Our study demonstrates the suitability of various methods for performing causal inference under several biologically plausible scenarios EwenCallaway(2017): “Biologists are likely to find that larger studies turn up more and more genetic variants – or “hits” - that have minuscule influences on disease” - Jonathan Pritchard, Stanford University matrix.
- 71. Multimodal Model Personalized precision medicine based on multiple measures with EHR mining. Example for diagnostics Example for management Example for treatment In 2015, aresearch groupatMountSinai Hospitalin New York wasinspiredtoapplydeeplearning to the hospital’s vast database of patient records. This data set features hundreds of variables on patients, drawn from their test results, doctor visits, and so on. The resulting program, which the researchers named Deep Patient, was trained using data from about 700,000 individuals, and whentestedonnewrecords,itprovedincrediblygoodatpredictingdisease. https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/ http://doi.org/10.1038/srep26094 https://github.com/greenelab/deep-review/issues/63 https://syncedreview.com/2017/02/26/deep-pa tient-improving-prognosis-with-electronic-h ealth-records-by-deep-learning/ on-demand.gputechconf.com
- 72. Multimodal Model Personalized precision medicine based on multiple measures with EHR mining. Example for diagnostics Example for management Example for treatment Thefuture ofhealth diagnostics. Currentdiagnosticsarebased on a “snapshot”in timeand limited datapoints.Inthefuture, largedatasetsacquired over timethroughconstantmonitoring will beanalyzed to establishbaselinesand trends,enabling preventativeinterventions. Partof thisfigurereusesa drawing previouslypublished in Swedish etal,2015.Copyright©2017ACM, Inc. Adapted fromSwedish T,Roesch K,LeeIK, Rastogi K, BernsteinS, RaskarR. EyeSelfie: Self directed eyealignmentusing reciprocaleyeboximaging. ACM TransGraph. 2015;34(4):58.39 KarinRoesch, TristanSwedish, and RameshRaskar (2017): “Automated retinalimaging andtrend analysis– atool forhealthmonitoring” https://dx.doi.org/10.2147/OPTH.S116265
- 73. Multimodal Model Personalized precision medicine based on multiple measures with EHR mining. Example for diagnostics Example for management Example for treatment Classificationofadvanced stagesofParkinson’s disease:translation into stratifiedtreatments JournalofNeuralTransmission August 2017, Volume124, Issue 8, pp1015–1027 RejkoKrüger, Jochen Klucken, Daniel Weiss, Lars Tönges, Pierre Kolber, Stefan Unterecker, Michael Lorrain, Horst Baas,Thomas Müller, Peter Riederer https://doi.org/10.1007/s00702-017-1707-x PrecisionMedicineinPediatric Oncology:Translating GenomicDiscoveriesinto OptimizedTherapies American Association forCancerResearch June9,2017 ThaiHoaTran, AvanthiTayiShah and Mignon L. Loh https://doi.org/10.1158/1078-0432.CCR-16-0115 Relative frequency of genomic alterations in neuroblastoma at diagnosiscompared with relapse
- 74. Usability Make it for the clinician, technician, eye- selfie taker the most easiest to use
- 75. Image Management Integrate deep learning algorithms to existing image management software The easiest for the end-user KideSystems– Optoflow https://www.kidesystems.com/optoflow/ DigisightPaxos https://www.digisight.net/ds/ Breaking Down Silos AcrossSpecialties Singapore National EyeCentre advanced itstechnology tocapture all images and fullyintegrate withitsEMR. http://go.merge.com/2017-Q1-OC-CS-Singapore-National-Eye-Centre_LP- Singapore-CS.html?utm_source=CS_Singapore https://www.aao.org/eyenet/article/image-manageme nt-systems-what-you-need-to-know
- 76. Interpretability Visualize what part of the image, 1D signal, the whole “disease network” is causing the deep learning system to flag up the patient in risk July–August,2017 Volume1,Issue4,Pages 322–327 CeciliaS.Lee,MD, DougM.Baughman,BS, Aaron Y.Lee,MD,MSCI DepartmentofOphthalmology, University ofWashingtonSchoolofMedicine,Seattle, Washington.http://dx.doi.org/10.1016/j.oret.2016.12.009 Anocclusiontest (ZeilerandFergus,2016) wasperformedtoidentifythe areas contributing most to the neural network's assigning the category of AMD. Ablank 20 × 20-pixel box was systematically moved across every possible position in theimageandtheprobabilitieswere recorded.The highestdropintheprobability represents the region of interest that contributed the highest importance to the deeplearningalgorithm. Examples of identification of pathology by the deep learning algorithm. Optical coherence tomography images showing age-related macular degeneration (AMD) pathology (A, B, C) are used as input images, and hotspots (D, E, F) are identified using an occlusion test from the deep learning algorithm. The intensity of the color is determined by the drop in the probability of being labeled AMD when occluded.
- 77. Interpretability Not just medical algorithms benefit from opening the “black box” Last month, a YouTubevideo of a conference talk in Berlin, shared widely among artificial-intelligence researchers, offered a possible answer. In the talk, NaftaliTishby, a computer scientist and neuroscientist from the Hebrew University of Jerusalem, presented evidence in support of a new theory explaining how deep learning works. Tishby argues that deep neural networks learn according to a procedure called the “information bottleneck,” which he and two collaborators firstdescribedinpurelytheoreticaltermsin1999. The idea is that a network rids noisy input data of extraneous details as if by squeezing the information through a bottleneck, retaining only the features most relevant to general concepts. Striking new computer experiments by Tishby and his student Ravid Shwartz-Ziv reveal how this squeezing procedure happensduringdeeplearning,atleastinthecasestheystudied. https://youtu.be/bLqJHjXihK8
- 78. Extra Resources Shallow introduction for Deep Learning Retinal Image Analysis https://www.slideshare.net/PetteriTeikariPhD/shallow-introduction-f or-deep-learning-retinal-image-analysis Shallow introduction for Deep Learning Retinal Image Analysis https://www.slideshare.net/PetteriTeikariPhD/artificial-intelligence- in-ophthalmology Shallow introduction for Deep Learning Retinal Image Analysis https://www.slideshare.net/PetteriTeikariPhD/datadriven-ophthalmol ogy Shallow introduction for Deep Learning Retinal Image Analysis https://www.slideshare.net/PetteriTeikariPhD/understanding-the-inve stors-medical-ai-startups