No abstract

Deep neural network models have produced significant results in solving various challenging tasks, including medical diagnostics. To increase the credibility of these black-box models in the eyes of doctors, it is necessary to focus on their explainability. Several papers have been published combining deep learning methods with selected types of explainability methods, usually aimed at analyzing medical image data, including ECG images. The ECG is specific because its image representation is only a secondary visualization of stream data from sensors. However, explainability methods for stream data are rarely investigated. Therefore, in this article we focus on the explainability of black-box models for stream data from 12-lead ECG. We designed and implemented a perturbation explainability method and verified it in a user study on a group of medical students with experience in ECG tagging in their final years of study. The results demonstrate the suitability of the proposed method, as well as the importance of including multiple data sources in the diagnostic process.

No abstract

Background: High prevalence and mortality of cardiovascular diseases (CVD) are global problems. Many countries focus their healthcare on secondary treatment, and primary prevention is in the background. Focusing on risk factors (RF) screening of CVD in personalized medicine has great potential for the future. We present a methodology for identification of important RF as well as the potentially new once. Materials and Methods: We worked with the dataset of patients hospitalized in the East Slovak Institute of Cardiovascular Diseases in Košice. The file contained 808 records, complete history, laboratory tests, ECG, echocardiography, and selective coronary angiography. We analyzed the importance of variables based on CART, Random forest, and Logistic regression algorithms (binary classification) and propose a new weighted agglomerative attribute importance metric. After selection of potentially important, but less known RF we re-deployed the CART algorithm on selected combinations of risk factors, while the target attribute was divided into six original classes corresponding to the severity of the coronarography finding.Results: Selected important variables based on the proposed metric are in accordance with known results, but also pointed to some potentially relevant RF such as fibrinogen. The experiments confirmed that fibrinogen might have the potential to help determine cardiovascular risk. However, its impact is debatable, but its potential increases if it is potentiated by factors other than HDL. We also concluded that higher HDL levels might have a cardio protective effect. Conclusions: Our study proposed original methodology for identification of important RF and in depth analysis of potentially interesting new RF. Results showed that fibrinogen could be one of the critical risk factors of cardiovascular diseases. Experimental results suggest that there may be others besides traditional RF of CVD, such as the fibrinogen and HDL levels we investigate. Still, studies with a larger patient population are needed to draw significant conclusions.

No abstract