Convolutional neural network-aided tuber segmentation in tuberous sclerosis complex patients correlates with...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Tuberous sclerosis complex (TSC) is characterized by cortical tubers, often leading to intractable epilepsy.
• Identifying the epileptogenic tuber is crucial for surgical planning but remains challenging.
• Current methods for tuber analysis can be invasive and time-consuming.

Purpose of the Study:

• To investigate the correlation between tuber characteristics and epileptogenesis in TSC.
• To develop and validate a deep learning model for automatic tuber segmentation and quantification.
• To assess the feasibility of using tuber burden as a biomarker for epilepsy in TSC.

Main Methods:

• Structural MRI and intracranial EEG data from 29 TSC subjects were analyzed.
• Logistic regression was used to correlate tuber statistics (volume, count, intensity) with epileptogenic zones.
• A neural network was trained for automatic tuber segmentation and burden quantification.

Main Results:

• Tuber volume and count per lobe, not voxel intensity, correlated with electrophysiological data.
• The lobe with the largest tuber volume matched the epileptic activity in 47.6% of subjects.
• The neural network achieved a sensitivity of 0.83 for tuber localization and accurate burden assessment.

Conclusions:

• Automatic tuber segmentation and burden quantification are feasible using deep learning.
• Tuber volume and count are significant indicators of epileptogenic zones in TSC.
• This approach offers potential for improved treatment and outcomes for TSC patients.