Group theoretic particle swarm optimization for multi-level threshold lung cancer image segmentation.

Area of Science:

• Medical Image Analysis
• Computational Intelligence
• Artificial Intelligence in Medicine

Background:

• Accurate medical image segmentation is crucial for computer-aided diagnosis, particularly in lung cancer detection.
• Manual segmentation is time-consuming and challenging for large datasets, necessitating automated solutions.
• Early tumor segmentation improves diagnosis, treatment planning, and patient survival rates.

Purpose of the Study:

• To develop a novel, automated method for medical image segmentation using evolutionary learning.
• To address the challenge of time-consuming manual segmentation in clinical routines.
• To improve the accuracy and efficiency of tumor region extraction in medical images.

Main Methods:

• Proposed Group Theoretic Particle Swarm Optimization (GTPSO) for multi-level threshold segmentation.
• GTPSO utilizes a novel search paradigm based on symmetric group theory, optimizing particle encoding, solution landscape, neighborhood movement, and swarm topology.
• Kapur's entropy was employed as the objective function to evaluate segmentation performance.

Main Results:

• GTPSO achieved superior performance compared to conventional metaheuristics for lung cancer image segmentation.
• Achieved a Kapur's entropy value of 9.07, representing a 16% improvement over the worst-case scenario.
• Demonstrated high accuracy with average evaluation metrics (Kappa, Precision, Recall, F1-measure, IoU, ROC) exceeding 90%, within an acceptable computational time of 173.730 seconds.

Conclusions:

• Group Theoretic Particle Swarm Optimization is an efficient method for segmenting medical images and extracting tumor tissues.
• The method balances diversification and intensification, leading to more accurate clinical diagnoses.
• The proposed approach holds significant potential medical value and clinical applicability.