Thesis Defence copy1
- 1. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 1SLIDE 1 1 Insights into fMR data using Machine Learning Presented By: Sonil Shrivastava(MT2014117) Supervisor : Prof Neelam Sinha
- 2. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 2SLIDE 2 2 Introduction to fMRIIntroduction to fMRI◆ ObjectiveObjective◆ Dataset DescriptionDataset Description◆ Introduction to RFEIntroduction to RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE ImplementationRFE Implementation :: • StarPlus DataStarPlus Data◦ Probid DataProbid Data◦ ◦ • ◦ Outline ROI AnalysisROI Analysis◦ Across SubjectsAcross Subjects Analysis -- 7 ROIsAnalysis -- 7 ROIs ◦ Across SubjectsAcross Subjects Analysis -- 3 ROIsAnalysis -- 3 ROIs ◦ ConclusionConclusion◆ ReferencesReferences◆
- 3. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 3SLIDE 3 3 Introduction To fMRIIntroduction To fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental Description : ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 4. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 4SLIDE 4 4 Introduction To fMRI Acquisition of the Blood Oxygen Level Dependent (BOLD) in fMRI fMRI -- functional Magnetic Resonance imaging ◆ Provides information about the functioning of the human brain ◆ ◆ Voxels and ROIs (Region of Interest).◆ voxels intensity --> f(x,y,z,t).◆
- 5. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 5SLIDE 5 5 Introduction -- fMRI◆ ObjectiveObjective◆ Dataset DescriptionDataset Description◆ Introduction -- RFE◆ Experimental Description : ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 6. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 6SLIDE 6 6 Objective Mapping of the brain corresponding to a cognitive state. ◆ Time stamp analysis.◆ Region of activation analysis.◆ Classification analysis across all the subjects. ◆
- 7. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 7SLIDE 7 7 Introduction -- fMRI◆ Objective◆ Dataset DescriptionDataset Description◆ Introduction -- RFE◆ Experimental Description : ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 8. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 8SLIDE 8 8 Dataset Description StarPlus Data 2 cognitive states(Picture and Sentence).◆ 54 trials, each trials of 27 seconds.◆
- 9. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 9SLIDE 9 9 Dataset Description StarPlus Data Data is dimension of 64x64x8◆ 5 subjects◆
- 10. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 10SLIDE 10 10 Dataset Description Probid Data 3 cognitive states (Pleasant , Unpleasant and Neutral). ◆ 5 subjects.◆ Dimension of 79x95x69, collected over 121 time points. ◆
- 11. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 11SLIDE 11 11 Voxel Elimination ? Data is high dimensional.◆ Minimum number of features 4500 for starPlus. ◆ Maximum number of features 219727219727 for probid. ◆ Solution --> RFE (Recursive feature Elimination) ◆
- 12. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 12SLIDE 12 12 Sample Data in fMRI Voxel 1 Voxel 2 Voxel 3 Voxel N Label value value value value Class 1 value value value value Class 1 value value value value Class 2 value value value value Class 2
- 13. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 13SLIDE 13 13 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction To RFEIntroduction To RFE◆ Experimental Description : ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 14. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 14SLIDE 14 14 Introduction To RFE Multivariate feature extraction algorithm.◆ Recursively eliminates irrelevant features.◆ SVM classifier --> for removing irrelevant voxels. ◆
- 15. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 15SLIDE 15 15 Introduction To RFE Flow Chart
- 16. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 16SLIDE 16 16 Introduction To RFE Algorithm
- 17. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 17SLIDE 17 17 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE ImplementationRFE Implementation :: • StarPlus DataStarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 18. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 18SLIDE 18 18 RFE implementation on starPlus Picture Vs Sentence Classfication Analysis for each subject◆ Each task -- 8 seconds -- 16 images◆ Number of features --> 16 * number of16 * number of voxelsvoxels ◆ 40 rows for picture and 40 rows for◆
- 19. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 19SLIDE 19 19 RFE implementation on starPlus Results Feature Extraction LevelFeature Extraction Level ClassificationAccuracyClassificationAccuracy
- 20. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 20SLIDE 20 20 RFE implementation on starPlus Results Feature Extraction LevelFeature Extraction Level ClassificationAccuracyClassificationAccuracy Results
- 21. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 21SLIDE 21 21 RFE implementation on starPlus Results Feature Extraction LevelFeature Extraction Level ClassificationAccuracyClassificationAccuracy Conclusion :Conclusion : As the feature extraction levels increases, accuracy also increases Results
- 22. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 22SLIDE 22 22 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE ImplementationRFE Implementation :: • StarPlus Data◦ Probid DataProbid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 23. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 23SLIDE 23 23 RFE implementation on Probid Pleasant Vs Unpleasant classification 5 subjects.◆ 50 rows for each of the task --> total 100 rows. ◆ Number of voxels : 219727.◆
- 24. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 24SLIDE 24 24 RFE on Pleasant Vs UnPleasant Results Feature Extraction LevelFeature Extraction Level ClassificationAccuracyClassificationAccuracy
- 25. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 25SLIDE 25 25 Feature Extraction LevelFeature Extraction Level ClassificationAccuracyClassificationAccuracy RFE on Pleasant Vs UnPleasant Conclusion :Conclusion : As the feature extraction levels increases, accuracy also increases
- 26. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 26SLIDE 26 26 RFE on Pleasant Vs Neutral Feature Extraction LevelFeature Extraction Level ClassificationAccuracyClassificationAccuracy Conclusion :Conclusion : As the feature extraction levels increases, accuracy also increases
- 27. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 27SLIDE 27 27 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE ImplementationRFE Implementation :: • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 28. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 28SLIDE 28 28 RFE Experiment for correctness Picture vs Sentence StarPlus Data.◆ Removed relevant voxels.◆ Accuracy is decreased as feature extraction level increases. ◆
- 29. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 29SLIDE 29 29 Reverse RFE Experiment for correctness Conclusion: More the relevant percentage of voxels removed, moreConclusion: More the relevant percentage of voxels removed, more sharply accuracy decreased.sharply accuracy decreased.
- 30. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 30SLIDE 30 30 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 31. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 31SLIDE 31 31 Time stamps analysis Important Time Stamps analysis. Objective
- 32. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 32SLIDE 32 32 Time stamps analysis Picture Vs Sentence data.◆ Analysis is done for individul subjects◆ Voxels from 7 ROIs◆ Accuracy measured for each time stamp◆ Procedure
- 33. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 33SLIDE 33 33 Results Time StampsTime Stamps ClassificationAccuracyClassificationAccuracy Time stamps analysis
- 34. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 34SLIDE 34 34 Voxels from 7-14 time stamps (3.5 sec to 7 sec) are more discriminating. ◆ This conforms to the concept of HRF (Heodynamic Response Function ) ◆ Discussion Time stamps analysis
- 35. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 35SLIDE 35 35 Results SubjectsSubjects ClassificationAccuracyClassificationAccuracy Time stamps analysis
- 36. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 36SLIDE 36 36 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI AnalysisROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 37. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 37SLIDE 37 37 ROI analysis Important Region of interest Analysis Objective
- 38. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 38SLIDE 38 38 ROI analysis Picture Vs Sentence data.◆ Voxels from 7 ROIs.◆ Analysis -- individual subjects.◆ Most discriminating Voxels are found from RFE. ◆ Distribution of these voxels are analyzed across the ROIs. ◆ Procedure
- 39. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 39SLIDE 39 39 Results Time StampsTime Stamps ClassificationAccuracyClassificationAccuracy ROI analysis
- 40. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 40SLIDE 40 40 3 ROIs (CALC, LIPL, LIPS) voxels are most discriminating. ◆ 70 % of the discriminating voxels are in this 3 regions. ◆ Discussion ROI analysis
- 41. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 41SLIDE 41 41 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Across SubjectsAcross Subjects Analysis -- 7 ROIsAnalysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ References◆
- 42. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 42SLIDE 42 42 Across all subjects.◆ Picture Vs Sentence data classification using RFE. ◆ Purpose of this analysis, to find class for generalized test data. ◆ Mean value across ROIs is taken.◆ Procedure Across Subjects analysis -- 7 ROIs
- 43. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 43SLIDE 43 43 Results Time StampsTime Stamps Across Subjects analysis -- 7 ROIs ClassificationAccuracyClassificationAccuracy
- 44. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 44SLIDE 44 44 Data is not classified accurately from 7 ROIs voxels, across all the subjects. ◆ Some irrelevant features needs to removed ◆ Discussion Across Subjects analysis -- 7 ROIs
- 45. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 45SLIDE 45 45 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental DescriptionExperimental Description :: ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Across SubjectsAcross Subjects Analysis -- 3 ROIsAnalysis -- 3 ROIs ◦ Conclusion◆ References◆
- 46. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 46SLIDE 46 46 Only 3 ROI's (CALC, LIPS ,LIPL) voxels taken.◆ Procedure Across Subjects analysis -- 3 ROIs
- 47. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 47SLIDE 47 47 Results Time StampsTime Stamps ClassificationAccuracyClassificationAccuracy Subjects analysis -- 3 ROIs
- 48. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 48SLIDE 48 48 Results Across Subjects analysis -- 3 ROIs Feature extraction levels, voxelsFeature extraction levels, voxels ClassificationAccuracyClassificationAccuracy
- 49. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 49SLIDE 49 49 Important voxels --> 3 ROIs (CALC, LIPS, LIPL)◆ Discussion Across Subjects analysis -- 3 ROIs
- 50. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 50SLIDE 50 50 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental Description : ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ ConclusionConclusion◆ References◆
- 51. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 51SLIDE 51 51 Conclusion Machine learning algorithm can be successfully utilized for fMR data analysis. ◆ Brain mapping corresponding to cognitive states established. ◆ 3 Important Regions are CALC, LIPS , LIPL in Picture vs Sentence analysis. ◆
- 52. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 52SLIDE 52 52 Introduction -- fMRI◆ Objective◆ Dataset Description◆ Introduction -- RFE◆ Experimental Description : ◆ RFE Implementation : • StarPlus Data◦ Probid Data◦ ◦ • ◦ Outline ROI Analysis◦ Subjects Analysis -- 7 ROIs ◦ Subjects Analysis -- 3 ROIs ◦ Conclusion◆ ReferencesReferences◆
- 53. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 53SLIDE 53 53 References De Martino, Federico and Valente, Giancarlo and Staeren, Noel and Ashburner, John and Goebel, Rainer and Formisano, Elia. Combining multivariate voxelCombining multivariate voxel selection and support vector machines for mapping andselection and support vector machines for mapping and classification of fMRI spatial patternsclassification of fMRI spatial patterns. Neuroimage, 43(1), 2008 ◆ Mitchell, Tom M and Hutchinson, Rebecca and Niculescu, Radu S and Pereira, Francisco and Wang, Xuerui and Just, Marcel and Newman, Sharlene. LearningLearning ◆
- 54. THESIS DEFENCE COPY1THESIS DEFENCE COPY1 sonilshrivastava22 SLIDE 54SLIDE 54 54 References Weblink. http://www.cs.cmu.edu/afs/cs.cmu.edu/project /theo-81/www/ StarPlus Data.StarPlus Data. ◆ Weblink. http://www.brainmap.co.uk/ PROBID data.PROBID data.◆