Faithfulness-Aware Visual-Evidence-Grounded Explanations for Cardiovascular Signal Image Classification

DOI

10.2991/978-94-6239-828-3_9How to use a DOI?

Keywords

XAI; Faithfulness-Aware; Classification; Cardiovascular Signal Image

Abstract

Cardiovascular signals such as electrocardiogram and photoplethysmography are increasingly analyzed using deep neural networks after being transformed into time-frequency images. Although convolutional models can achieve strong classification performance, their predictions remain difficult to interpret, and natural language explanations generated by language models may sound plausible without being faithful to the visual evidence used by the classifier. This paper presents a faithfulness-aware explanation pipeline for cardiovascular signal image classification. The system first transforms cardiovascular signals into scalogram images and classifies them using a lightweight convolutional backbone. Grad-CAM is then used to extract structured visual evidence, including the dominant temporal-frequency region, activation area ratio, peak activation score, predicted class, and confidence. These evidence fields condition both template-based and lightweight language-model explanations. We evaluate the system using classification metrics and explanation reliability metrics, including evidence support rate, hallucination rate, label agreement, and evidence coverage. The proposed evidence-grounded explanation is expected to reduce unsupported clinical statements compared with unconstrained language-model explanations while preserving interpretability for non-expert users. This study is positioned as an explainable AI prototype, not a clinical diagnostic system. Our research shows that CNN can classify scalogram images with high accuracy, demonstrating good performance. Meanwhile, Grad-cam also provides good heatmaps. Future research should focus on using more advanced and complex LLMs.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Muhammad Alfaiz Khalifah Alamsyah
AU  - Novanto Yudistira
AU  - Wayan Firdaus Mahmudy
PY  - 2026
DA  - 2026/11/04
TI  - Faithfulness-Aware Visual-Evidence-Grounded Explanations for Cardiovascular Signal Image Classification
BT  - Proceedings of the 2026 11th International Conference on Sustainable Information Engineering and Technology (SIET 2026)
PB  - Atlantis Press
SP  - 88
EP  - 106
SN  - 2589-4900
UR  - https://doi.org/10.2991/978-94-6239-828-3_9
DO  - 10.2991/978-94-6239-828-3_9
ID  - Alamsyah2026
ER  -