A Comparative Study of Explainable Artificial Intelligence Algorithms in Local Interpretable Model-Agnostic based Pneumonia Prediction

Deep learning from AI field has exhibited great success in various tasks related to medical imaging, including pneumonia detection from chest X-ray data. However, the lack of transparency and black-box nature of these models raises concerns about their reliability along with trustworthiness in clinical environment. Recently, explainable artificial intelligence (XAI) methods have attracted research communities to make the models trust-worthy and interpretable. In this context, this paper makes both visual and quantitative comparative study of super-pixel based explainable algorithms which are integrated to the best performing deep learning model from VGG16, LENET, INCEPTION, SPINAL, and ALEXNET using Local Interpretable Model-Agnostic Explanations (LIME) tool. Three super-pixel algorithms, Quick-shift, SLIC, and Quick-Shift with noise, are evaluated using standard performance metrics. Simulation results show that Quick-Shift with Noise algorithm performs optimally with Jaccard co-efficient value of 0.9679.