Fuzzy production rules and deep learning neural networks: explicable artificial intelligence 2.0 for the diagnosis of coronary stenosis
Автор: Trofimov Yu.V., Semashko V.S., Muravyov I.P., Kuznetsov E.M., Averkin A.N.
Журнал: Сетевое научное издание «Системный анализ в науке и образовании» @journal-sanse
Рубрика: Моделирование и анализ данных
Статья в выпуске: 2, 2025 года.
Бесплатный доступ
The article proposes a multi-level approach to diagnosing coronary stenosis based on deep learning and fuzzy logic. It addresses problems of insufficient labeled data through additional annotation of the CADICA dataset. An algorithm has been implemented including vessel segmentation using a modified U-Net network and CRF, augmented with XAI methods (Grad-CAM, LIME, Score-CAM). A neuro-fuzzy module ANFIS transforms model`s activations into rules. The approach provides high segmentation accuracy (Dice ≈ 0.84; IoU ≈ 0.78) and diagnostic reliability even in subtle pathology cases. Results confirm in-creased expert trust due to integration of explainable AI mechanisms.
Echocardiography, image segmentation, deep neural networks, explicable artificial intelligence, neuro-fuzzy systems, coronary stenoses
Короткий адрес: https://sciup.org/14133180
IDR: 14133180