Convolutional neural network with group theory and random selection particle swarm optimizer for enhancing cancer...

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Computational Biology

Background:

• Convolutional Neural Networks (CNNs) excel in medical image analysis for diagnosis and prognosis.
• CNN performance is highly dependent on optimal parameter tuning.
• Existing optimization methods may not fully leverage model structure and parameter correlations.

Purpose of the Study:

• To propose a novel self-tuning CNN model incorporating a metaheuristic-based optimizer.
• To introduce Group Theory and Random Selection-based Particle Swarm Optimization (GTRS-PSO) for efficient CNN parameter optimization.
• To enhance accuracy and minimize errors in medical image classification tasks.

Main Methods:

• Developed GTRS-PSO by extracting insights from symmetric model structures and parameter correlations.
• Implemented hierarchical partitioning of the parameter space with four operators.
• Utilized a random selection strategy for parameter updates within partitions.
• Applied the GTRS-PSO optimized CNN to breast and lung cancer radiology datasets.

Main Results:

• GTRS-PSO achieved superior performance in cancer image classification compared to other optimization algorithms.
• The proposed method demonstrated effectiveness, particularly with datasets exhibiting symmetric properties.
• Optimized CNN models using GTRS-PSO showed enhanced accuracy in distinguishing between cancerous and non-cancerous images.

Conclusions:

• The novel GTRS-PSO optimizer offers a significant advancement in tuning CNNs for medical image analysis.
• This approach effectively addresses the sensitivity of CNNs to parameter settings.
• GTRS-PSO holds promise for improving diagnostic accuracy in radiology, especially for conditions with inherent data symmetries.