A Framework for Interactive Medical Image Segmentation Using Optimized Swarm Intelligence with Convolutional Neural...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Advancements in image processing technology significantly impact medical imaging applications.
• Accurate segmentation of diagnostic images is crucial for medical research.
• There is a need for effective deep learning-based segmentation techniques for rapid and precise identification of regions of interest.

Purpose of the Study:

• To propose and evaluate a pipeline for medical image segmentation using Convolutional Neural Networks (CNNs) and Swarm Intelligence (SI).
• To compare the performance of six different segmentation modules, including traditional clustering algorithms and their hybridizations with CNNs and PSO.

Main Methods:

• A pipeline integrating CNNs and SI for image segmentation was developed.
• Six modules were evaluated: Fuzzy C-means (FCM), K-means, FCM with Particle Swarm Optimization (PSO), K-means with PSO, FCM with CNN, and K-means with CNN.
• Experiments were conducted on diverse medical image datasets (MRI, dermoscopic, microscopic, CT) with varying data subset sizes (50 to 2000 images).

Main Results:

• The K-means with CNN module demonstrated superior performance compared to other evaluated methods.
• Achieved a segmentation accuracy of 96.45% with an average processing time of 9.09 seconds.
• Performance was evaluated across multiple medical imaging modalities and dataset sizes.

Conclusions:

• The proposed K-means with CNN approach offers an effective solution for medical image segmentation.
• Hybrid deep learning and swarm intelligence methods show significant promise for improving diagnostic accuracy and efficiency in medical imaging.
• Further research can explore advanced CNN architectures and SI algorithms for enhanced medical image analysis.