Comparative analysis of machine learning algorithms for predicting stereotactic coordinates of the centromedian...

Area of Science:

• Neurosurgery
• Medical Imaging
• Machine Learning

Background:

• Deep brain stimulation (DBS) of the centromedian nucleus (CM) is a key treatment for refractory epilepsy and other neurological disorders.
• Accurate CM targeting is difficult due to poor MRI visualization, hindering surgical outcomes and accessibility.

Purpose of the Study:

• Develop and validate machine learning (ML) models to predict CM coordinates using conventional T1-weighted MRI.
• Enhance the precision and accessibility of CM-DBS procedures.

Main Methods:

• Four ML models (LR, KNN, SVR, DNN) were trained on 100 MRI scans from healthy individuals.
• Models predicted CM stereotactic coordinates using readily identifiable MRI landmarks.
• Validation was performed on 20 epilepsy patients undergoing CM-DBS.

Main Results:

• The deep neural network (DNN) model achieved the highest accuracy, with mean errors of 0.88 mm (healthy) and 1.12 mm (epilepsy).
• Other models (LR, SVR, KNN) showed comparable performance but with higher error rates.

Conclusions:

• ML models, particularly DNNs, can reliably predict CM coordinates from standard T1-weighted MRI.
• This method reduces reliance on advanced imaging, making CM-DBS more feasible in diverse clinical settings.