Pneumonia Detector

An ensemble model for pneumonia detection that achieves 95.51% accuracy and 96.44 % F1 score on the ChestXRay17 dataset.

Setup

git clone git@github.com:rafaelglikis/pneumonia-detector.git
cd pneumonia-detector
pip install -r requirements.txt

Download and extract dataset

mkdir dataset
wget https://data.mendeley.com/datasets/rscbjbr9sj/2/files/f12eaf6d-6023-432f-acc9-80c9d7393433/ChestXRay2017.zip?dl=1 -O dataset/dataset.zip
unzip dataset/dataset.zip -d dataset/
# Also remove unnecessary files
rm -r dataset/__MACOSX
rm dataset/dataset.zip
rm dataset/chest_xray/test/.DS_Store
rm dataset/chest_xray/train/.DS_Store

Download ensemble models

• Download models.
• Extract
• Move the contents of the ensemble directory to the ensemble directory of this project.

The contents of the ensemble directory should be:

ensemble
├── inception_v3_transfer_20200725-123618
├── resnet50_v2_transfer_20200808-134834
├── vgg16_v3_transfer_20200726-124845
└── xception_20200811-013119

Usage

usage: pneumdet.py [-h]
                   [--train {inception,vgg16,resnet50,densenet121,xception,mobilenet}]
                   [--evaluate EVALUATE [EVALUATE ...]]
                   [--ensemble {evaluate}]

Detect pneumonia from chest x rays.

optional arguments:
  -h, --help            show this help message and exit
  --train {inception,vgg16,resnet50,densenet121,xception,mobilenet}
                        Train a model.
  --evaluate EVALUATE [EVALUATE ...]
                        Evaluate trained model.
  --ensemble {evaluate}
                        Use ensemble.

Examples

Train inception model

 python pneumdet.py --train inception

Evaluate ensemble/inception_v3_transfer_20200725-123618 model

python pneumdet.py --evaluate ensemble/inception_v3_transfer_20200725-123618 

Evaluate all models in the ensemble directory

 python pneumdet.py --evaluate ensemble/*

Evaluate the ensemble created

 python pneumdet.py --ensemble evaluate