Multiple Lung Disease Classification Using Fine Tuned Transfer Learning: An Explainable AI Approach

Multiple Lung Disease Classification is an important problem for health care stakeholders. Tuberculosis, COVID-19, and Pneumonia are found to be severe lung diseases. All these diseases commonly have many symptoms like cough, fever, fatigue, and other breathing difficulties. Hence, a healthcare practitioner finds it very challenging to identify which lung disease the patient belongs to. Hence, it is important to develop a Machine Learning (ML) for classifying an X Ray image into 3 categories namely Tuberculosis, COVID 19, and Pneumonia. So, we developed a ML model consists of transfer learning using pre-trained Residual Networks (Res Net) classifier, reported with a Balanced Classification Accuracy (BCA) of 98.2 %, precision and recall of 98.3 %, F1 Score of 0.98. We compared the proposed method with that of other transfer learning methods such as InceptionV3, DenseNet 201, and LeNet using BCA and F1 Score. Then, we applied Local Interpretable Model-Agnostic Explanations (LIME) based explainability to the developed model for gaining a better understanding of the developed model.