Enhancing Pneumonia Detection: CNN Interpretability with LIME and SHAP

Authors:

Abstract
A bacterial or viral infection of the lungs can cause pneumonia, one of the dangerous and potentially fatal illnesses that can have dire repercussions in a short amount of time. Therefore, a key component of a successful treatment plan is an early diagnosis. Therefore, a sophisticated and automated system that can diagnose chest X-rays and make the process of diagnosing pneumonia easier for both specialists and novices is required. This research aims to create a CNN model that will help with the accurate classification of pneumonia. In this work, we have presented our Deep Learning method for the classification challenge, which is taught using modified images through several pre-processing stages. With an overall accuracy of 93.30%, we were able to classify X-ray images of pneumonia using a custom CNN model. Our proposed model is able to precisely detect pneumonia from X-ray images with amazing accuracy and loss. Furthermore, we employed the LIME and SHAP tools of the XAI technique to generate a noteworthy conclusion to persuade medical practitioners.

This publication has 15 references indexed in Scilit: