Convolutional neural network (CNN) configuration using a learning automaton model for neonatal brain image segmentation

Iran Sarafraz ¹, Hamed Agahi ¹, Azar Mahmoodzadeh ¹

⁰Department of Electrical Engineering, Shiraz Branch, Islamic Azad University, Shiraz, Iran.

Plos One +

|

January 17, 2025

View abstract on PubMed

Summary

This study introduces an adaptive method for configuring convolutional neural networks (CNNs) for neonatal brain image segmentation. The approach optimizes CNNs using reinforcement learning, improving segmentation accuracy for infant MRI scans.

Area Of Science

Medical Imaging
Artificial Intelligence
Neuroscience

Background

Convolutional Neural Networks (CNNs) are effective for brain image segmentation but require specific adaptations for neonatal brain images due to developmental differences.
Optimal CNN architecture and parameters are crucial for accurate neonatal brain image segmentation.

Purpose Of The Study

To present an adaptive method for automatic CNN configuration tailored for neonatal brain image segmentation.
To optimize CNN hyperparameters, including filter dimensions and pooling functions, using reinforcement learning and a Look-Ahead (LA) model.

Main Methods

An encoder-decoder CNN structure was employed.
Reinforcement learning and an LA model were utilized to determine optimal hyperparameters (filter size, length, width, pooling type).
Segmentation quality was evaluated using DICE and Average Symmetric Surface Distance (ASD) metrics.

Main Results

The proposed adaptive method successfully configured CNNs for neonatal brain image segmentation.
The optimized CNNs achieved higher quality and accuracy in segmenting neonatal brain images (NBI) compared to previous methods.
Evaluation on an infant MRI database demonstrated the effectiveness of the approach.

Conclusions

The adaptive CNN configuration method significantly enhances the accuracy and quality of neonatal brain image segmentation.
This approach provides a robust framework for optimizing deep learning models in specialized medical imaging applications.
The findings suggest a promising direction for improving automated analysis of infant brain development through advanced AI techniques.