Automatic diagnosis of pathological myopia from heterogeneous biomedical data

Area of Science:

• Ophthalmology
• Biomedical Informatics
• Medical Imaging
• Genetics

Background:

• Pathological myopia is a leading cause of global blindness, particularly in Asia.
• It involves retinal degeneration, potentially leading to irreversible vision loss if untreated.
• Current diagnostic methods often rely on single data types, limiting comprehensive assessment.

Purpose of the Study:

• To develop and evaluate a computer-aided diagnosis framework for pathological myopia.
• To enhance diagnostic accuracy by integrating heterogeneous biomedical data.
• To improve early detection and disease management for pathological myopia.

Main Methods:

• Proposed a novel framework: Pathological Myopia diagnosis through Biomedical and Image Informatics (PM-BMII).
• Utilized multiple kernel learning (MKL) to fuse diverse data types.
• Evaluated the framework using data from 2,258 subjects, including demographic, clinical, retinal imaging, and genotyping data.

Main Results:

• The PM-BMII framework achieved an Area Under the Curve (AUC) of 0.888.
• This significantly outperformed methods using only demographic/clinical data (AUC 0.607, +46.3%), genotyping data (AUC 0.774, +14.7%), or imaging data (AUC 0.852, +4.2%).
• Results demonstrate the effectiveness of integrating heterogeneous data for improved pathological myopia diagnosis.

Conclusions:

• The PM-BMII framework offers a feasible and effective approach for diagnosing pathological myopia.
• Integrating multiple data sources provides a more holistic understanding of disease risk factors.
• This approach holds promise for timely intervention and better management of pathological myopia.