Multiple Machine Learning Classifiers to identify various types of lung disease
Multiple Machine Learning Classifiers to identify various types of lung disease: pneumonia, COVID-19. Used VGG-16 and ResNet50 to build pneumonia and COVID classifiers. I wanted to see whether transfer learning could be applied to a problem such as medical imaging while being pretrained on ImageNet, a general computer vision task. It performed quite well. Pneumonia classification with 92%, 98%, and 95% accuracy on test batches. COVID-19 validation accuracy was above 96% at end. The same ensemble that was used on COVID-19 was used on lung cancer to classify adenocarcinoma(aca), sqaumous cell carcinoma(scc), and healthy lung tissue. A 96% accuracy was reached towards the end of training.
Example image of pneumonia x-ray:
Example image of COVID-19 CT Scan:
Example image of Adenocarcinoma(aca):
Example image of Sqaumous Cell Carcinoma(scc):
Example image of healthy lung tissue:
Architecture of pneumonia model: VGG-16(VGG-16 architecture pretrained on ImageNet with modified top) see VGG-16 architecture at bottom above ResNet50.
Architecture of COVID-19 Model(Ensemble of VGG-16 and ResNet50 models pretrained on ImageNet below) picture to be updated:
Below is the graph of the final training epoch for the COVID-19 classifier:
ResNet Architecture:
VGG Architecture:
Datasets:
https://www.kaggle.com/andrewmvd/covid19-ct-scans
https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia
https://www.kaggle.com/andrewmvd/lung-and-colon-cancer-histopathological-images(colon data discarded)