- Information
- AI Chat
Was this document helpful?
Practical - Diagnosis of heart disease via cnns
Course: Convolutional Neural Networks for Visual Recognition (CS 231n)
183 Documents
Students shared 183 documents in this course
University: Stanford University
Was this document helpful?
Diagnosis of Heart Disease via CNNs
Kaicheng Wang
Stanford University
kwang2@stanford.edu
Yanyang Kong
Stanford University
yanyangk@stanford.edu
Abstract
Our project predicts volume of heart by 2D MRI mea-
surement. Combining pre-trained VGG [13] and self-
trained networks, we build our Convolutional Neural Net-
works (CNNs) for prediction. Specific preprocessing meth-
ods are designed for our messy data. CNNs in various
depths and regularization strength are tried for best vali-
dation result. In this Kaggle Challenge contest, our model
beats the baseline CNN structure written by Marko Jocic
[6].
1. Introduction
Using MRI data to measure cardiac end-systolic (VS)
and end-diastolic (VD) volumes (i.e., the size of one cham-
ber of the heart at the beginning and middle of each heart-
beat) (fig[1]) and then deriving the ejection fraction (EF)
of heart is a standard process to assess the heart’s squeez-
ing ability. Declining EF is a key indicator of heart disease.
However, the current process is manual and slow. The cardi-
ologist could spend up to 20 minutes with one patient. Con-
sidering the huge amount of patients with potential heart
failure, quick automatic measurement will help doctors to
diagnose heart conditions more efficiently. 1
EF = 100 ·VD−VS
VD
Convolutional Neural Networks (CNNs) have been
proved remarkably effective on neuroimaging data. As a
powerful visual model, CNNs can yield many interesting
hierarchies of features, which can be used to classifica-
tion and segmentation. Contemporary CNN models like
AlexNet[8], VGG-Net [13], GoogLeNet[14], ResNet[5]
have been used and transferred to learn representations by
fine-tunning weights in many different tasks.
In our project, we are going to apply deep Convolutional
Neural Networks to predict the end-systolic volume VSand
1This project is the Second Annual Data Science Bowl Challenge prob-
lem: https://www.kaggle.com/c/second-annual-data-science-bowl
end-diastolic volume VDfrom MRI time series data in dif-
ferent axis views (the planes of slice).
(a) end-systolic volume (b) end-distolic volume
Figure 1: Plots of of end-systolic, end-distolic volumes in
one cardiac circle, circled in red curve
1.1. Problem Statement
Given one patient’s several (10 axes in average) MRI
measurements in different axes (sax 5,, 2ch 12, 4ch 32,
etc), we are going to predict his or her left ventricle’s end-
distole volume and end-systole volume in one cardiac cycle.
Each measurement contains 30 times series images in one
cardiac cycle.
We aim to build two deep CNN regression models to
predict these two volumes separately. The overview of our
problem is shown in fig [2].
Figure 2: Using 30 MRIs during one cardiac cycle from
different axis views to predict VSand VD
1
Students also viewed
- CS231n 2016-2017
- Practical - Imagenet classification with deep convolutional neural networks
- Summary - Gradient based learning applied to document recognition
- Practical - Multi-view 3d pose estimation from single depth images
- Summary - Real-time depth estimation from 2d images
- Practical - Random search for hyper-parameter optimization