An improved 3D-UNet-based brain hippocampus segmentation model based on MR images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Accurate hippocampal delineation from MRI is vital for diagnosing neurosystemic diseases.
• Current methods for hippocampus segmentation face challenges in speed and accuracy.

Purpose of the Study:

• To develop an automatic hippocampus segmentation method using 3D-UNet.
• To improve the accuracy and efficiency of hippocampus segmentation in MRI.

Main Methods:

• A pixel-level semantic segmentation approach using 3D-UNet was employed.
• A filling technique was integrated into the segmentation network.
• The model was trained and validated on 200 MRI datasets, with comparisons against VNet, SegResNet, and UNetR.

Main Results:

• The 3D-UNet model with the filling technique achieved a Dice score of 0.7989 ± 0.0398 and mIoU of 0.6669 ± 0.0540.
• Performance improved with increased input image size, reaching a Dice score of 0.8674 ± 0.0257 for 96x96x96 resolution.
• The enhanced model demonstrated superior segmentation accuracy compared to original networks and other models.

Conclusions:

• The proposed hippocampus segmentation model with the filling technique outperforms existing methods.
• This advancement enhances diagnostic analysis efficiency for neurosystemic diseases.