Automating stroke lesion segmentation in brain images using a multi-model multi-path convolutional neural network Yunzhe.

Slides:



Advertisements
Similar presentations
Variability of LIDC Panel Segmentations and Soft Segmentation of Lung Nodules Presented by Stephen Siena and Olga Zinoveva.
Advertisements

1 Automated Feature Abstraction of the fMRI Signal using Neural Network Clustering Techniques Stefan Niculescu and Tom Mitchell Siemens Medical Solutions,
Faculty of Computer Science © 2007 Information Theoretic Measures: Object Segmentation and Tracking CMPUT 615 Nilanjan Ray.
NA-MIC National Alliance for Medical Image Computing Competitive Evaluation & Validation of Segmentation Methods Martin Styner, UNC NA-MIC.
Correction of bias field artifact in T1w MR images of the thigh and calf muscles Correction de contraste d’images IRM Projet A.
Developing outcome prediction models for acute intracerebral hemorrhage patients: evaluation of a Support Vector Machine based method A. Jakab 1, L. Lánczi.
Accuracy, Reliability, and Validity of Freesurfer Measurements David H. Salat
Kim HS Introduction considering that the amount of MRI data to analyze in present-day clinical trials is often on the order of hundreds or.
Radboud University Medical Center, Nijmegen, Netherlands
Automatic Lung Cancer Diagnosis from CT Scans (Week 3)
Cancer Metastases Classification in Histological Whole Slide Images
Combining Models Foundations of Algorithms and Machine Learning (CS60020), IIT KGP, 2017: Indrajit Bhattacharya.
When deep learning meets object detection: Introduction to two technologies: SSD and YOLO Wenchi Ma.
Deep Learning for Dual-Energy X-Ray
CNN-RNN: A Unified Framework for Multi-label Image Classification
Learning to Compare Image Patches via Convolutional Neural Networks
Validation and Evaluation of Algorithms
Automatic Lung Cancer Diagnosis from CT Scans (Week 1)
Debesh Jha and Kwon Goo-Rak
Machine Learning for Big Data
Object Detection based on Segment Masks
Sentence Modeling Representation of sentences is the heart of Natural Language Processing A sentence model is a representation and analysis of semantic.
Journal Club M Havaei, et al. Université de Sherbrooke, Canada
Data Mining, Neural Network and Genetic Programming
HyperNetworks Engın denız usta
Convolutional Neural Fabrics by Shreyas Saxena, Jakob Verbeek
MR images analysis of glioma
Applications of Deep Learning and how to get started with implementation of deep learning Presentation By : Manaswi Advisor : Dr.Chinmay.
Robust Lung Nodule Classification using 2
Compositional Human Pose Regression
Effective Connectivity
Synthesis of X-ray Projections via Deep Learning
Lecture 5 Smaller Network: CNN
SemanticFusion: Dense 3D Semantic Mapping with Convolutional Neural Networks Paper by John McCormac, Ankur Handa, Andrew Davison, and Stefan Leutenegger.
Wellcome Trust Centre for Neuroimaging University College London
Bird-species Recognition Using Convolutional Neural Network
Normalized Cut Loss for Weakly-supervised CNN Segmentation
Principles of using neural networks for predicting molecular traits from DNA sequence Principles of using neural networks for predicting molecular traits.
Outline Announcement Texture modeling - continued Some remarks
Volume 10, Issue 6, Pages (June 2018)
Paper Presentation Aryeh Zapinsky
Toward improved document classification and retrieval
Jason furmanek Kris murphy IBM
Deep neural networks (DNNs) can successfully identify, count, and describe animals in camera-trap images. Deep neural networks (DNNs) can successfully.
RGB-D Image for Scene Recognition by Jiaqi Guo
AlexNet+Max Pooling MIL AlexNet+Label Assign. MIL
Road Traffic Sign Recognition
المشرف د.يــــاســـــــــر فـــــــؤاد By: ahmed badrealldeen
EFnet: an added value of multi-model simulation
Effective Connectivity
Anatomical Measures John Ashburner
边缘检测年度进展概述 Ming-Ming Cheng Media Computing Lab, Nankai University
Daily Warm Up 08/26/2015 – 08/27/2015 List, in order, the steps of the scientific method. The independent variable is the thing that is _______________.
Deep Learning for Automated Contouring of Primary Tumor Volumes by MRI for Nasopharyngeal Carcinoma In this multicenter evaluation, artificial intelligence.
Abnormally Detection
CSC 578 Neural Networks and Deep Learning
Automatic Handwriting Generation
U-Net: Convolutional Network for Segmentation
VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION
Weekly Learning Alex Omar Ruiz Irene.
End-to-End Facial Alignment and Recognition
Pose Estimation in hockey videos using convolutional neural networks
MS-SEG Challenge: Results and Analysis on Testing Data
Recent Developments on Super-Resolution
Identification of Vertebral Fractures by Convolutional Neural Networks to Predict Nonvertebral and Hip Fractures: A Registry-based Cohort Study of Dual.
Directional Occlusion with Neural Network
An introduction to Machine Learning (ML)
Presented By: Firas Gerges (fg92)
Shengcong Chen, Changxing Ding, Minfeng Liu 2018
Deep CNN for breast cancer histology Image Analysis
Presentation transcript:

Automating stroke lesion segmentation in brain images using a multi-model multi-path convolutional neural network Yunzhe

Why automatic lesion identification? Reduce time to identify lesion Manual lesion identification can take 4 to 10 hours whereas automatic takes minutes (Wilke et. al. NeuroImage 2011) Human inter-rater variability Independent raters agreement ranges from 67% to 73% Fiez et. al. Human Brain Mapping 2000 Sook-Lei et. al. Scientific Data 2018

What does our system do? Our system takes in three views of the MRI scan (each normalized in three different ways). We can see the human identified lesion in each view. The bottom images contain the prediction given by our system overlaid on the same input images.

Our system

Our system

First phase

First phase’s results

3D CNN Post-processing

Before 3D CNN Post Processing After 3D CNN Post Processing

Experimental performance study MRI images from three different sources Kessler Foundation (N = 25) Medical College of Wisconsin (N = 20) ATLAS public dataset (N = 54) All datasets contain human curated lesions that serve as ground truth in our experiments and in training our models Do cross-study and cross-validation Measure accuracy with the commonly used Dice coefficient

5-fold cross-validation on all data

Big vs small lesions Large lesions are generally easier to identify than smaller ones Define small lesions as below 10,000 pixels in size and large otherwise We perform better than previous methods on both big and small lesions: Statistically better than the popular DeepMedic on small lesions Statistically better than UNet on large lesions

Big vs small lesions Raincloud plots of accuracies of our system

Thank you