计算机科学
人工智能
核(代数)
机制(生物学)
模式识别(心理学)
卷积(计算机科学)
机器学习
人工神经网络
算法
支持向量机
特征(语言学)
卷积神经网络
作者
Urousha Jeetun,Maleika Heenaye-Mamodekhan
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-07-01
卷期号:: 1-6
标识
DOI:10.1109/nextcomp70574.2026.11708182
摘要
Cardiac disease classification from Cardiac Magnetic Resonance Imaging (CMR) is a challenging diagnostic task, complicated by the morphological similarity between pathology classes and the limited availability of annotated imaging data. This paper proposes a classification system based on a Selective Kernel Convolutional (SKC) network with embedded spatial attention mechanisms, evaluated on the ACDC benchmark dataset comprising 150 CMR scans across five diagnostic categories. Two parallel CNN branches with complementary receptive fields extract multi-scale features, with a spatial attention module embedded within each branch to focus on diagnostically relevant regions. Grad-CAM visualisation is integrated post-hoc to generate saliency maps highlighting the anatomical regions most influential to each classification decision. The proposed system achieves 94% accuracy with precision and recall exceeding 80% across all five classes, achieving comparable performance to prior methods on this benchmark while offering a segmentation-free pipeline with improved interpretability for clinical adoption.
科研通智能强力驱动
Strongly Powered by AbleSci AI