Large Model Driven Multi-Granularity Medical Image Analysis: A Fuzzy Logic-Guided Framework.

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Image analysis

Background:

• Medical image analysis requires advanced computational methods to address complexity and uncertainty in pathological structures.
• Existing transformer-based approaches face challenges in capturing multi-scale pathological features effectively.

Purpose of the Study:

• To develop a large model-driven framework, ULVM-MG, integrating fuzzy logic and transformers for multi-granularity medical image analysis.
• To enhance diagnostic accuracy by processing pathological images at coarse, medium, and fine levels.

Main Methods:

• The ULVM-MG framework utilizes a multi-granularity feature extraction strategy.
• A fuzzy-guided cross-attention mechanism is employed to focus on diagnostically significant regions while retaining contextual information.
• The approach integrates fuzzy logic principles with transformer architectures.

Main Results:

• ULVM-MG achieved superior performance on histopathological datasets (LC25000 and NCT) compared to state-of-the-art transformer methods.
• Accuracy rates of 98.76% (LC25000) and 97.34% (NCT) were recorded, outperforming baselines by 1.61% and 2.17%, respectively.
• The framework demonstrated particular strength in differentiating similar tissue types and classifying benign versus malignant cases.

Conclusions:

• The proposed ULVM-MG framework effectively enhances medical image analysis through multi-granularity processing and fuzzy uncertainty modeling.
• The integration of fuzzy logic and transformers offers significant improvements in accuracy and diagnostic capabilities for pathological images.
• ULVM-MG represents a substantial advancement in computational pathology, offering a more systematic and accurate approach to image interpretation.