Explainable AI Based Malaria Detection Using Lightweight CNN

A customized lightweight Convolutional Neural Network (CNN) for the fastest malaria detection from RBC images with explainability is presented in this study. Accuracy, precision, recall, f1-score and number of model parameters are considered for evaluating the proposed method. It is called lightweight as it has a small number of parameters (0.17 M). A small number of model parameters can reduce the processing time generally. It's performance is compared with the well-known transfer-learning (TL) model's performance. Additionally, the proposed method is also compared with state-of-the-arts (SOTA) models. In both the cases, the proposed method performed similarly or better in every performance criterion as mentioned earlier. The proposed CNN model achieves 99.45%, 99.75%, 99.17% and 99.46% scores on accuracy, precision, recall and f1-score respectively. Explainable Artificial Intelligence (XAI) tool ‘SHAP’ (Shapley Additive Explanations) is used to explain the decision of the proposed model. The results show that the proposed lightweight CNN is an efficient model in terms of performance parameters as mentioned for malaria detection.