A comparative study of deep learning based Dadashi-Tazehozi rd2669 - PowerPoint PPT Presentation

A comparative study of deep learning based methods for MRI image processing Robert A comparative study of deep learning based Dadashi-Tazehozi rd2669 methods for MRI image processing Introduction Articles Motivation Medical Robert Dadashi-Tazehozi background rd2669 Neurological Diseases MRI Image processing pipeline Department of Computer Science Datasets Columbia University Preprocessing Random Forest for Classification Deep Learning for Computer Vision and Natural Results Language Processing EECS 6894

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative Articles study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 Deep Learning for Cerebellar Ataxia Classification and Functional Score Regression Introduction Articles Zhen Yang, Shenghua Zhong, Aaron Carass, Sarah H. Ying, Motivation and Jerry L. Prince Medical background Johns Hopkins University Neurological Diseases MRI Image processing Deep Learning of Image Features from Unlabeled Data pipeline Datasets for Multiple Sclerosis Lesion Segmentation Preprocessing Random Forest for Youngjin Yoo, Tom Brosch, Anthony Traboulsee, David K.B. Classification Results Li, and Roger Tam University of British Columbia

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative Motivation study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 ◮ MRI data Introduction ◮ Deep Learning + Machine Learning Articles Motivation ◮ Different goals and barriers while same data type, hence Medical background a comparative study Neurological Diseases - Data description MRI Image processing - Preprocessing pipeline - Methods and Algorithms used Datasets Preprocessing - Results Random Forest for Classification Results ◮ Project: Diabetic Retinopathy Detection ◮ Personal reasons

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative Ataxia study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 Overview A neuro-degenerative disease Introduction Articles - Affects the cerebellum Motivation - Symptoms: lack of muscular coordination Medical background Neurological Diseases https://www.youtube.com/watch?v=5eBwn22Bnio MRI Image processing pipeline Goals: Datasets Preprocessing Classify different types of Ataxia: HC, SCA2, SCA6, AT Random Forest for Classification Results Quantify functional loss based on structural change

A comparative Multiple sclerosis study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 Overview A inflammatory disease Introduction Articles - Brain cells damaged: Demyelination Motivation - Symptoms: Mental, Physical, Psychatric troubles Medical background - Cause: Genetic? Environmental factors? Neurological Diseases MRI Image processing https://www.youtube.com/watch?v=qgySDmRRzxY pipeline Datasets Preprocessing Goals: Random Forest for Classification Results Automatic segmentation of lesionned areas

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative MRI imaging study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 Introduction ◮ Human body composed of small magnets Articles Motivation ◮ Magnets aligned then excited by pulses Medical background ◮ Magnets go back to their lowest energy state, Neurological Diseases MRI electromagnetic waves emitted Image processing ◮ Processing of these waves enable to reconstruct 3D pipeline Datasets structure, differentiate muscles tissues from fat, white Preprocessing Random Forest for Classification matter from grey matter in the brain Results

A comparative MRI imaging study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 Figure: Initial state Introduction Articles Motivation Medical background Neurological Diseases MRI Image processing pipeline Datasets Figure: Excitation Preprocessing Random Forest for Classification Results Figure: Energy emission

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative Datasets study of deep learning based Ataxia methods for MRI image processing ◮ 168 scans Robert ◮ 3D Images projected on 9 plans (Resulting in 32 * 32 Dadashi-Tazehozi rd2669 images) Introduction Articles Motivation Medical background Neurological Diseases MRI Image processing pipeline Datasets Not enough labelled data ⇒ Interest of generating Preprocessing Random Forest for synthetic data Classification Results Multiple Sclerosis ◮ 1450 scans ◮ Resolution 256 * 256 * 50 ◮ Only 100 scans segmented Lot of unlabelled data ⇒ Interest of unsupervised methods

A comparative Outline study of deep learning based methods for MRI image processing Robert Introduction Dadashi-Tazehozi rd2669 Articles Motivation Introduction Articles Motivation Medical Medical background background Neurological Diseases Neurological Diseases MRI MRI Image processing pipeline Datasets Preprocessing Image processing pipeline Random Forest for Classification Datasets Results Preprocessing Random Forest for Classification Results

A comparative Preprocessing study of deep learning based methods for MRI image processing Ataxia Robert Dadashi-Tazehozi ◮ 3D Images projected on 9 plans (Resulting in 32 * 32 rd2669 images) Introduction Articles ◮ Generate translated and rotated images Motivation ◮ Reduce dimensions using a Stacked Auto-Encoder for Medical background each plan Neurological Diseases MRI Image processing Multiple Sclerosis pipeline Datasets ◮ Images of resolution 256 * 256 * 50 Preprocessing Random Forest for ◮ Patches 9 * 9 * 3 (low-scale features) Classification Results ⇒ Train Restricted Boltzmann Machines for feature extraction ◮ Patches 15 * 15 * 5 (high-scale features) ⇒ Train Deep Belief Network for feature extraction

A comparative Stacked Auto-Encoder study of deep learning based methods for MRI image processing Robert Dadashi-Tazehozi rd2669 Introduction Articles Motivation Medical background Neurological Diseases MRI Image processing pipeline ◮ Target vector is the output Datasets Preprocessing ◮ Along the way, compression of the data Random Forest for Classification Results ◮ Trained layer after layer (greedy)

A comparative Restricted Boltzman Machines study of deep learning based methods for MRI image processing Robert ◮ Visible and Hidden Units Dadashi-Tazehozi rd2669 Introduction Articles Motivation Medical background Neurological Diseases MRI Image processing pipeline Datasets ◮ Energy Based Method Preprocessing Random Forest for Classification Results E ( v , h ) = − v T Wh − a T v − b T h P ( v , h ) = 1 Z e − E ( v , h )