Towards MRI-Only Mandibular Resection Planning: CT-like Bone Segmentation from Routine T1 MRI Images Using Deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Current head and neck cancer surgical planning relies on both CT and MRI, necessitating complex image registration and exposing patients to radiation.
• This dual-modality approach presents challenges in workflow efficiency and resource utilization.

Purpose of the Study:

• To develop and validate a deep learning model for accurate bone segmentation directly from T1-weighted MRI scans.
• To enable MRI-only virtual surgical planning, thereby streamlining the process for head and neck oncology.

Main Methods:

• A deep neural network was trained using CT-derived segmentations (mandible, cranium, inferior alveolar nerve) as ground truth.
• A dataset of 100 patients with paired CT and MRI scans was utilized, with MRI data preprocessed to match CT voxel size and alignment.
• The model was trained on 80 cases and evaluated on 20 cases using Dice similarity coefficient, Intersection over Union (IoU), precision, and recall.

Main Results:

• The deep learning model achieved high accuracy in bone segmentation, with a mean Dice score of 0.86 and IoU of 0.76.
• Precision and recall were both 0.86, indicating robust performance.
• Surface deviation analysis revealed minimal differences (median 0.21-0.30 mm) between CT- and MRI-derived bone models.

Conclusions:

• Accurate, CT-comparable bone models can be generated from standard T1-weighted MRI scans using deep learning.
• This validates the feasibility of an MRI-only approach for virtual surgical planning in head and neck oncology.
• The findings support reduced radiation exposure and improved efficiency in surgical planning workflows.