Development of accurate human head models for personalized electromagnetic dosimetry using deep learning.

Area of Science:

• Medical imaging analysis
• Computational electromagnetics
• Deep learning applications

Background:

• Personalized human head models are crucial for electromagnetic dosimetry, including electrostimulation optimization and safety assessments.
• Current methods for generating head models via MRI segmentation are time-consuming and require expertise, hindering accurate electric field computation in specific brain regions.
• Existing deep learning approaches primarily focus on brain tissue segmentation, neglecting other critical head structures for dosimetry.

Purpose of the Study:

• To propose a novel convolutional neural network architecture, ForkNet, for comprehensive segmentation of whole human head structures.
• To enable the generation of accurate personalized head models essential for evaluating electric field distribution in the brain.
• To improve the efficiency and accuracy of head model creation for applications like transcranial magnetic stimulation.

Main Methods:

• Development of ForkNet, a new convolutional neural network architecture.
• Application of ForkNet for segmenting diverse human head anatomical tissues from medical images.
• Validation of ForkNet-generated head models against manually segmented models.

Main Results:

• ForkNet successfully performs segmentation of whole human head structures.
• Generated head models show strong agreement with manually segmented models in intra-scanner tasks.
• The proposed network facilitates the creation of personalized head models for electromagnetic dosimetry.

Conclusions:

• ForkNet offers an efficient and accurate deep learning solution for whole human head segmentation.
• The developed method supports the generation of personalized head models for advanced electromagnetic dosimetry applications.
• This work advances the capability to evaluate electric field distribution in the brain, particularly during transcranial magnetic stimulation.