BackgroundDiffusion tensor imaging (DTI) is a commonly utilized pre-surgical tractography technique. Despite widespread use, DTI suffers from several critical limitations. These include an inability to replicate crossing fibers and a low angular-resolution, affecting quality of results. More advanced, non-tensor methods have been devised to address DTI’s shortcomings, but they remain clinically underutilized due to lack of awareness, logistical and cost factors.ObjectiveNath et al. (2020) described a method of transforming DTI data into non-tensor high-resolution data, suitable for tractography, using a deep learning technique. This study aims to apply this technique to real-life tumor cases.MethodsThe deep learning model utilizes a residual convolutional neural network architecture to yield a spherical harmonic representation of the diffusion-weighted MR signal. The model was trained using normal subject data. DTI data from clinical cases were utilized for testing: Subject 1 had a right-sided anaplastic oligodendroglioma. Subject 2 had a right-sided glioblastoma. We conducted deterministic fiber tractography on both the DTI data and the post-processed deep learning algorithm datasets.ResultsGenerally, all tracts generated using the deep learning algorithm dataset were qualitatively and quantitatively (in terms of tract volume) superior than those created with DTI data. This was true for both test cases.ConclusionsWe successfully utilized a deep learning technique to convert standard DTI data into data capable of high-angular resolution tractography. This method dispenses with specialized hardware or dedicated acquisition protocols. It presents an economical and logistically feasible method for increasing access to high definition tractography imaging clinically.