A Hybrid Neuroevolutionary Approach to the Design of Convolutional Neural Networks for 2D and 3D Medical Image Segmentation

The evolution of Convolutional Neural Networks (CNNs) has revolutionized medical image segmentation, yet designing optimal architectures remains a challenge. In this paper, we introduce a hybrid evolutionary algorithm that advances the design of CNNs for medical image segmentation. By integrating Cartesian Genetic Programming with Simulated Annealing, our approach efficiently explores the architectural design space, yielding CNN architectures that perform robustly across several medical imaging tasks. Our model demonstrated competitive performance on the DRIVE and Medical Segmentation Decathlon datasets, achieving Dice Similarity Coefficients up to 0.828, and excelling in computational efficiency with reduced GFLOP values. Notably, our algorithm’s design eliminates the need for extensive pre-training, reducing computational overhead and enhancing its suitability for resource-constrained environments.