4Our study aimed to develop an explanatory method for predicting Coronary Artery Disease (CAD) classification using spect images. As we all know, deep neural networks usually consist of many layers connected to each other through interlocking network nodes. Even if we check the classes and describe their relationships, it is difficult to understand entirely how active neural networks make predictions. Therefore, deep learning is still considered a "Black box". Existing XAI (eXplainable Artificial Intelligence) approach can provide insights into the inside of a Deep Learning model allowing for transparency and interpretation. Our previous research helps doctors diagnose the CAD of patients by developing deep learning models using a multi-stage transfer learning framework. The model achieved 0.955 accuracy, 0.932 AUC, 0.944 sensitivity, and 0.889 specificity, showing effective performance. Our dataset includes 218 SPECT images from 218 imported patients collected at 108 Hospital in Hanoi, Vietnam. In this paper, We propose an explainable Deep Learning framework using three popular XAI approaches: LIME, GradCam, and RISE. These XAI approaches are effective tools for interpreting the prediction of deep learning models. We evaluate the effectiveness of the interpretation by visualizing the explained regions and using improved deletion and insertion with a threshold limit suitable for Binary Classification. The experiment results show that our model effectively diagnoses CAD and provides medical interpretation. Furthermore, the proposed method for evaluating the deletion and insertion metrics is considered more efficient for binary classification than the traditional metrics.

No abstract

Artificial intelligence applications, especially deep learning in medical imaging, have gained much attention in recent years. With the computer's aid, Coronary artery disease (CAD) -one of the most dangerous cardiovascular diseasesis diagnosed effectively without human interference and efforts. A lot of research involving predicting CAD from Myocardial Perfusion SPECT has been conducted and given impressive results. However, all existing methods detect whether there is a disease or not. They do not provide information about which obstructive areas are (mainly in the left anterior descending artery (LAD), left circumflex artery (LCx), and right coronary artery (RCA) territories) that result in CAD. To further diagnose CAD, we develop new classifiers to solve a multilabel classification problem with the highest accuracy and area under the receiver operating characteristics curve (AUC) when compared to different methods. Our proposed method is based on transfer learning to extract features from Myocardial Perfusion SPECT Polar Maps and a novel stack of sub-classifiers to detect particularly obstructive areas. We evaluated our methods with eight hundred and one obstructive images from a database of patients referred to a hospital from 2017 to 2019.

No abstract