Artificial intelligence model for tumoral clinical decision support systems

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Comparative diagnostics leverage existing medical data to aid in new patient evaluations.
• Current AI models for medical image analysis often require complex tumor segmentation.
• Costly and difficult segmentation processes limit the accessibility of advanced diagnostic tools.

Purpose of the Study:

• To develop an AI system for retrieving similar brain tumor cases for enhanced diagnostic accuracy.
• To generate accurate medical image representations focusing on patient-specific features and pathologies.
• To eliminate the need for tumor segmentation by using binary information for enriched image descriptors.

Main Methods:

• An AI model was developed to detect patient features and recommend similar cases from a database.
• The system balances healthy and abnormal feature representation to improve generalization.
• The AI utilizes binary information to create enriched image descriptors, bypassing segmentation.

Main Results:

• The proposed AI architecture achieved a Dice coefficient of 0.474 in both tumoral and healthy regions, outperforming prior studies.
• The model effectively extracts and combines anatomical and pathological features from brain Magnetic Resonances (MRs).
• State-of-the-art results were achieved with reduced training costs due to reliance on less expensive label information.

Conclusions:

• The presented AI architecture offers significant potential for improving diagnostic efficiency and accuracy in brain tumor treatment.
• Further research is warranted to explore the broader applicability and optimization of this novel approach.
• This AI-assisted image retrieval system promises to reduce costs and enhance patient care by acting as a transparent support tool.