pneumonia-CNN

Scanning chest x-rays for pneumonia using a deep convolutional network

Example

A healthy and diseased x-ray next to each other.

Training results

Accuracy vs Batch & Loss vs Batch

Accuracy vs Epoch & Loss vs Epoch

0.9375 final validation accuracy

0.9828 final training accuracy

Architecture

Layer Type	Activation	Parameters
Conv 2D	RelU	3x3 kernel 32 filters
Max Pool 2D	None	2x2 stride
Conv 2D	RelU	3x3 kernel 32 filters
Max Pool 2D	None	2x2 stride
Conv 2D	RelU	3x3 kernel 64 filters
Max Pool 2D	None	2x2 stride
Conv 2D	RelU	3x3 kernel 64 filters
Max Pool 2D	None	2x2 stride
Flatten	None	None
Fully Connected	RelU	128 units
Dropout	None	0.5 rate
Fully Connected	RelU	128 units
Dropout	None	0.5 rate
Fully Connected	Softmax	2 units

Data

The dataset can be found here

-- Kermany, Daniel; Zhang, Kang; Goldbaum, Michael (2018), “Labeled Optical Coherence Tomography (OCT) and Chest X-Ray Images for Classification”, Mendeley Data, v2 http://dx.doi.org/10.17632/rscbjbr9sj.2

CC BY 4.0