Lungs Disease Classification Using an Self Attention Based Neural Network: A Local Agnostic Explainable Approach

Understanding the future potential candidates for lung
disease classification poses a great challenge for physicians.
If they can understand the future lungs disease
patients from their Chest X-ray images, it would help
them in proper design of medication strategies. It helps
them to reduce their overburdening during physical examination
of their patients. Physician often face difficulty
in accurately finding the proper chest disease. However,
an explainable neural network enabled architecture
can help in better diagnosis of lung diseases. Hence,
this study proposes an attention-based Convolutional
Neural Network and consequently followed by a local
agnostic explanation method for better understanding
what made the artificial intelligence model to predict the
image belongs to either Bacterial Tuberculosis, Pneumonia,
Normal, Viral Pneumonia. We achieved a BCA
of 84% on the testing set. Moreover, an explainable AI
based local agnostic framework is used to understand
which portion of the lungs make a class distinguishable.