Effective Approaches to Fetal Brain Segmentation in MRI and Gestational Age Estimation by Utilizing a Multiview Deep...

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Accurate quantitative analysis of fetal brain development is crucial for understanding neurodevelopment in both healthy and abnormal fetuses.
• Automatic multi-tissue fetal brain segmentation is essential for this quantitative analysis.

Purpose of the Study:

• To propose an effective, end-to-end automatic multi-tissue fetal brain segmentation model (IRMMNET).
• To develop and evaluate methods for predicting gestational age (GA) using the proposed model and radiomics features.
• To assess the generalization capabilities of the proposed methods on other medical imaging tasks.

Main Methods:

• Developed IRMMNET incorporating inception residual encoder blocks and dense spatial attention blocks for multi-scale feature extraction and reduced parameters.
• Implemented three gestational age prediction methods: 3D autoencoder, radiomics features, and IRMMNET encoder.
• Experimented on 80 fetal brain MRI volumes (20-33 weeks GA), manually segmented into seven tissue categories.

Main Results:

• IRMMNET achieved a Dice score of 0.791±0.18 for fetal brain segmentation, surpassing state-of-the-art methods.
• Radiomics-based gestational age prediction yielded the best performance with a Root Mean Square Error (RMSE) of 1.42.
• Demonstrated successful generalization to head and neck tumor segmentation and patient survival prediction.

Conclusions:

• The proposed IRMMNET model offers an effective solution for automatic multi-tissue fetal brain segmentation.
• Radiomics features provide a robust approach for accurate gestational age prediction from fetal brain MRI.
• The developed methods show promise for broader applications in medical image analysis and prediction.