OCIF: automatically learning the optimized clinical information fusion method for computer-aided diagnosis tasks.

Area of Science:

• Medical image analysis
• Artificial intelligence in healthcare
• Computational neuroscience

Background:

• Accurate computer-aided diagnosis (CAD) requires fusing image features from neural networks with clinical information.
• Integrating low-dimensional clinical information (LDCF) with high-dimensional network features (HDNF) presents a significant challenge in CAD.
• Existing methods struggle to efficiently combine these disparate data types for improved diagnostic performance.

Purpose of the Study:

• To introduce a novel network search framework, OCIF (Optimized Clinical Information Fusion), for effective LDCF integration with HDNF.
• To provide an optimized solution for LDCF fusion and efficient dimensionality reduction in HDNF within CAD systems.
• To enhance diagnostic accuracy in CAD by leveraging both clinical and image-derived features.

Main Methods:

• OCIF utilizes Gaussian process optimization to explore optimal neural network architectures, including fully connected layers, neuron counts, activation functions, dropout rates, and clinical information inclusion.
• Transfer learning is employed within OCIF to reduce the parameter space and accelerate the search process.
• The framework's effectiveness was evaluated using three popular end-to-end overall survival (OS) time prediction models on a relevant dataset.

Main Results:

• Applying OCIF to a standard CAD neural network significantly improved classification accuracy.
• Experiments on the 2020 BRATS dataset demonstrated OCIF's satisfactory performance in OS time prediction, achieving an accuracy of 0.684, precision of 0.735, recall of 0.684, and F1-score of 0.675.
• The results confirm OCIF's capability to enhance diagnostic performance in medical classification tasks.

Conclusions:

• OCIF effectively integrates clinical information with network features, enhancing diagnostic accuracy in computer-aided diagnosis.
• The framework demonstrates the potential for significant improvements in CAD accuracy by leveraging multimodal data.
• OCIF shows promise for extension to various other medical classification tasks, offering a versatile approach to data fusion.