<p>In recent years, the application of machine learning in medical fields, including image diagnosis, has been advancing. However, the black-box nature of machine learning has raised issues such as the assurance of interpretability. In the field of image recognition, various explainable AI methods such as Grad-CAM have been developed. However, detailed information such as geometric features in the images is lost, leaving issues of reliability and persuasiveness unresolved. This paper proposes an explainable AI using topological data analysis(TDA), which is widely used in medical fields. One of the key features of TDA is the ability to identify the corresponding area of the original input image from the extracted features through inverse problem analysis. Using chest CT image data collected at Hokkaido University Hospital, we were able to demonstrate the qualitative and quantitative superiority of the visualization method over the conventional method (Grad-CAM) in terms of squared loss mutual information value (conventional method 0.1568, proposed method 0.2933). Furthermore, we constructed a more understandable visualization method by decomposing the visualization results according to the degree of feature inclusion in each class.</p>