Detecting Pneumonia with Dual Path Ensembled Deep Learning Models on Chest X-ray Images

A 96% accurate dual-path AI using lung-segmented and full chest X-rays with ResNet18, DenseNet121, and InceptionV3 ensembles outperforms single models to aid rapid pneumonia diagnosis, especially where it kills a child every 43 seconds.

Paper: Detecting Pneumonia with Dual-Path Ensembled Deep Learning Models with Lung Segmentation on Chest X-ray Images

Code:

Lung Segmentation (UNet with VGG11 Encoder): Code
Pneumonia Detection (Averaged Ensembled): Code

Datasets:

CRD: Chest X-Ray Images with Lung Segmented Masks: Dataset Link
PneumoniaCXR: Normal, Pneumonia, NoOpacityNoNormal: Dataset Link