A Modified Genetic Algorithm Based Encoder-Decoder Network for Brain Tumor Detection in 3D Medical Images
Year: 2025
Type: article
Abstract: Timely detection of brain tumors is critical for improving treatment effectiveness and patient outcomes. The advancement of 3D medical imaging, particularly Magnetic Resonance Imaging (MRI), has significantly enhanced diagnostic accuracy by offering detailed anatomical views beyond traditional 2D methods. However, detecting brain tumors in MRI images poses significant challenges due to their diverse locations, shapes, and intensity variations. Moreover, manual design and fine-tuning of hyperparameters for 3D Convolutional Neural Network (CNN) models are inherently complex tasks, necessitating expert domain knowledge. The large size of 3D images further compounds this complexity, requiring substantial memory resources and increasing model size. To address these challenges, this paper introduces a novel method based on Neural Architecture Search (NAS) to design optimal 3D Encoder-Decoder (ED) CNNs for enhancing brain tumor detection performance while simultaneously reducing network size. A comprehensive search space is designed with various block structures such as transformers, squeeze-and-excitation modules, and efficientNet blocks, alongside various activation functions and normalization layers, along with essential training hyperparameters like loss function and optimizer. Furthermore, a modified Genetic Algorithm (E-GA) is introduced, featuring modified selection and population update operations, to explore the search space efficiently. The E-GA identifies optimal parameters, enabling the construction of 3D ED networks that explore trade-offs between detection performance and model size, forming a set of Pareto-optimal solutions. Evaluation of the proposed method on two benchmark brain tumor datasets demonstrates superior performance and reduced model size compared to state-of-the-art methods, underscoring its potential to improve 3D medical image analysis and enhance patient outcomes.
DOI: 10.1145/3774319
Language: English
Subfield: Neurology
Field: Neuroscience
Domain: Life Sciences
Open Access status: bronze
OpenAlex ID: W4415736116