This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Atherosclerosis through Hierarchical Explainable Neural Network Analysis
Created by
Haebom
Author
Irsyad Adam, Steven Swee, Erika Yilin, Ethan Ji, William Speier, Dean Wang, Alex Bui, Wei Wang, Karol Watson, Peipei Ping
Outline
This paper develops a hierarchical graph neural network framework to address the personalized classification problem of atherosclerosis. By leveraging two feature modes—patient clinical characteristics and individual patient molecular data—we overcome the Limitations (lack of consistency and understanding of cohort-wide features) inherent in existing graph-based disease classification methods. A novel framework, ATHENA (Atherosclerosis Through Hierarchical Explainable Neural Network Analysis), builds a hierarchical network representation through integrated mode learning and optimizes patient-specific molecular fingerprints reflecting individual omics data to enhance consistency with cohort-wide patterns. Experimental results on 391 patients demonstrated up to 13% improvement in area under the curve (AUC) and up to 20% improvement in F1 score. XAI-based subnetwork clustering enables mechanistically informed patient subtype discovery, enhancing personalized intervention strategies and improving atherosclerosis progression prediction and clinical outcome management.
Takeaways, Limitations
•
Takeaways:
◦
Improved personalized classification performance of atherosclerosis using hierarchical graph neural networks (up to 13% AUC, up to 20% F1 score increase).
◦
Discovery of patient-specific molecular fingerprints and subtypes through integration of clinical features and molecular data.
◦
Enhancing mechanistic understanding and enhancing personalized intervention strategies through XAI-based subnetwork clustering.
◦
Improving prediction of atherosclerosis progression and management of clinical outcomes.
•
Limitations:
◦
The size of the dataset used in this study (391 people) may be relatively small.
◦
Further studies are needed to determine generalizability to other types of atherosclerosis or other diseases.
◦
The computational cost and complexity of the ATHENA framework need to be evaluated.
◦
Further validation of the explanatory power and reliability of the XAI method used is needed.
