Big data and AI in a radiologist's reading room
1 like657 views
The document explores trends in radiology emphasizing the shift from analog to digital processes, integration of AI/ML in imaging, and the importance of quality measurement within radiological practices. It discusses challenges such as staff shortages, error management, and the necessity for standardization and interoperability in imaging data. Additionally, it highlights the need for motivation and effective change management in the transition towards advanced healthcare automation and AI-driven solutions.
Big data and AI in a radiologist's reading room
- 1. www.medradiology.moscow Professor Sergey Morozov Eketerina Guseva Big data and AI in radiologist’s reading room Atlas – June 16, 2017
- 2. www.medradiology.moscow Agenda • Radiology trends • How to measure imaging quality • Moscow URIS • How to manage changes • AI/ML for imaging data analysis
- 5. www.medradiology.moscow Radiology trends • From modalities (CT, MRI, X-Ray, etc.) to clinical sections • From analog to digital • From a dark room to open space • From jack-of-all-trades to labor division • From separated to integrated clinical facilities • From art to craft and skills • From unique services to mass production
- 9. www.medradiology.moscow MRI CT US NM Digitizer PACS DB Server Short-term archive Long-term archive Diagnostic workstations Web server Clinical Web- clients RIS server RIS client PACS RIS RIS clients Printing DicomIzer Mammo X-ray Cloud-based imaging All formats image sharing
- 11. www.medradiology.moscow Familiar problems? • Shortage of qualified staff • Low motivation • “Cherry picking” principle • High staff turnover • Slow implementation of new methods France, Imaging management, 2009
- 12. www.medradiology.moscow Quality problem • IoM, USA: 100 000 deaths/year of medical errors • Russia: …..? 12 Original report According to CT, the tumor of the middle pole of the left kidney. Cyst of the left kidney. Multiple myomas in the uterus. Peer-review Cyst of the left kidney. Multiple myomas in the uterus. False positive CT diagnosis of renal cancer
- 15. www.medradiology.moscow Quality Domains § Appropriateness § Diagnostic accuracy § Safety § Actionable reporting § Communication Measuring Radiologist’s Value Add
- 17. www.medradiology.moscow Moscow Radiology nowadays 166 CT 101 MRI2147 RG 27 SPECT+PET193 MMG 2634 eq 12,5 MM 707 K 641 K 222 K 215 K 14,3 MM exams (per 2015 year) +22%+26%-1% +7%+8%
- 18. www.medradiology.moscow 77 med centers 62 CT 40 MRI140 radiologists + 22 experts RPC Medical Radiology 26 mammo United RIS (Radiology Information Service) 3 PET/CT Administrative department: + ORACLE dashboard Teaching and consulting dpt: + Teleaudit + Moodle CME 15 X-Ray5 fluorograph > 700 000 studies
- 20. www.medradiology.moscow Data availability and structuring • Mapping of insurance codes, imaging services nomenclature (Min. of Health) • Translation: – “any 6-hours interval with at least 1 study uploaded to URIS” = ”shift” – ”absence of images from a scanner at 9am” = possibly out-of-order • Pilot automation of Statistics Report #30 – successful for CT and MRI • Develop unified methodology of calculating numbers of studies, procedures, visits, referrals, etc.
- 22. www.medradiology.moscow Teleaudit of CT, MRI, mammo Reports Ratings Bonuses Webinars CME Random 5% selection of studies Anonymized remote viewing Expert’s detection of errors
- 24. www.medradiology.moscow 9,4 22,6 20,5 47,5 4,6 14 10,4 69 0 10 20 30 40 50 60 70 80 Клинически значимые расхождения Клинически не значимые расхождения Замечания общего характера Полное соответствие I кв. 2016 I кв. 2017 Out-patient medical centers Teleaudit results 24 Clinically significant Not clinically significant General remarks and recommx No remarks
- 25. www.medradiology.moscow Reports and ratings TAT = report turn-around time Healthcare facility 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Presentation to CMO Head of radiology department Healthcare Department Avg. report turnaround time decrease by 80% (from 53 to 8 hours) TAT CT QIV’15 TAT CT QI’16 CHANGE 25
- 27. www.medradiology.moscow 27 Competition among CMOs of hospitals Positive motivation to achieve the effectiveness goals Managing results by ratings ГБУЗ "ГП № 64 ДЗМ" ГБУЗ "ГП № 66 ДЗМ" ГБУЗ "ГП № 69 ДЗМ" ГБУЗ "КДЦ № 2 ДЗМ" ГБУЗ "ГП № 195 ДЗМ" ГБУЗ "ГП № 209 ДЗМ" ГБУЗ "ГП № 212 ДЗМ" ГБУЗ "ГП № 8 ДЗМ" ГБУЗ "КДЦ № 4 ДЗМ" ГБУЗ "ГП № 45 ДЗМ" ГБУЗ "ГП № 6 ДЗМ"
- 29. www.medradiology.moscow Target audience • Radiologists • Clinicians • Technicians 03-2016 05-2017 0 50 100 150 200 250 300 350 Jan – Apr 2017 • 45 webinars • 6885 attendees from 20 regions of Russia Webinars of RPC Medical Radiology I.Trofimenko i.trofimenko@rpcmr.org.ru Free webinars of RPCMR
- 32. www.medradiology.moscow Healthcare automatization • Общий технический потенциал ≈ 36% • Питание пациентов, распределение медикаментов > 75% • Сбор информации от пациентов > 50% • Анализ снимков ≈ 30% • Медсестра < 30% • Стоматолог-гигиенист ≈ 13% 32morozov@npcmr.ru
- 34. www.medradiology.moscow #AIHealth DIAGNOSTICS AND RISK ANALYTICS DRAW MOST DEALS IN 2016 Imaging & diagnostics dominated 2015, and is at par with insights & risk analytics so far this year. Deals to lifestyle management & monitoring picked up in 2015. Virtual assistants, and AI startups in mental health still have room for further investment activity. www.cbinsights.com 16 AI IN HEALTHCARE: CATEGORY HEATMAP 2012-2016 (as of 8/30/2016) No. of deals Min Max CATEGORY 2012 2013 2014 2015 2016YTD Insights & Risk Analytics Lifestyle Management & Monitoring Imaging & Diagnostics Drug Discovery Emergengy Room & Hospital Management Wearables Virtual Assistants Mental Health Research Nutrition Miscellaneous www.cbinsights.com Investments in AI/ML for imaging 34 morozov@npcmr.ru Enlitiс.com develops AI/ML • X-ray/CT/MR • CXR AUC 0.92 • Radiologist’s AUC 0.5-0.72 Automatic lung nodes detection, measurement, classification, and reporting
- 35. www.medradiology.moscow Adverse events • PACS è less communications between radiology and clinical departments • Mobile images è commoditization • Shared workflow è confidentiality breaches • Turf battles between radiologists and cardiologists, orthopedic surgeons, neurologists • CAD/AI/ML è decrease of radiologist’s role • «Digital jungle aboriginals» è new principles of teaching and work
- 37. www.medradiology.moscow AI/ML business model – yet to be defined
- 40. www.medradiology.moscow Special challenges in working with physicians • High need for autonomy • Sensitivity to criticism • Perfectionistic & compulsive • Want to direct; resist control • Arrogance often indicates low self esteem From Barbara Linney & Kent Neff MD, ACPE; also Valliant, Gabbard, Dawson
- 42. www.medradiology.moscow How to motivate? Change management: – Team involvement – Motivation – Emotions – Habit – Persistence
- 43. www.medradiology.moscow Perspectives for 2017-2019 • Further healthcare standardization • Competitive advantages – Service, results availability, quality, expert’s availability • Expert as a service – Telereporting – B2B • AI/ML-based services for mass imaging analysis (lung Ca screening, mammoDx) – Automation of routine tasks of doctors: overall potential of 36%