Machine Learning-Based Synthetic Computed Tomography Derived From Temporal Bone Magnetic Resonance Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Otology

Background:

• Current otologic diagnostics and presurgical planning often require both MRI for soft tissues and CT for bone visualization.
• A single, radiation-free imaging modality for visualizing both soft and bony tissues is highly desirable.

Purpose of the Study:

• To develop and assess a machine learning algorithm for generating synthetic CT images from head MRI scans.
• To evaluate the feasibility of using synthetic CT for otologic applications.

Main Methods:

• A machine learning algorithm was trained on 67 paired MRI and CT scans.
• Synthetic CT images were generated from MRI data.
• 15 synthetic CT scans were clinically evaluated by ENT surgeons and radiologists for accuracy, landmark conspicuity, and suitability.

Main Results:

• Synthetic CT images demonstrated sufficient geometric and radiodensity accuracy compared to true CT.
• Landmark conspicuity was comparable, though bone thickness was sometimes overestimated and ossicles were often not depicted.
• Most synthetic CT scans were suitable for localization, navigation, and surgical planning, but not for primary diagnosis.

Conclusions:

• Machine learning can generate synthetic CT-like images from MRI, enabling visualization of bony and soft tissues in a single, radiation-free session.
• Synthetic CT images are suitable for localization and preoperative planning in otologic procedures, including estimating mastoid pneumatization.