Enhancing Hippocampal Subfield Visualization Through Deep Learning Reconstructed MRI Scans

Background/Objectives: Assessing hippocampal pathology in epilepsy is challenging, and improving diagnostic accuracy can benefit from deep learning image reconstruction, standardized imaging protocols, and advanced post-processing methods. This study compares T2 TSE DRB (Deep Resolve Boost) sequences with standard T2 TSE sequences for hippocampal segmentation and volumetry using FreeSurfer, focusing on how DRB affects image acquisition time without compromising diagnostic accuracy. Methods: FreeSurfer (version 7.4.1) was used to segment hippocampal subregions in 36 subjects (mean age of 39 ± 14 years; 21 males, 15 females) using both T2 TSE DRB and T2 TSE sequences. The segmented volumes were compared with a two-tailed t-test, and pathological volume differences were assessed using z-values based on a 95% confidence interval (-2 < z < 2). Results: Overall hippocampal segment volumes were identical between sequences. However, significant volume differences were noted in the CA1-Body (p = 0.003), CA4-Body (p = 0.002), and whole hippocampal body (p = 0.012) in the right hippocampus. Despite these differences, the low effect sizes suggest DRB sequences are comparable to conventional sequences. Additionally, DRB reduced image acquisition time by 61%. Z-scores identified pathological volume changes between the left and right hippocampus in individual subjects. Conclusions: T2 TSE DRB sequences are non-inferior to conventional T2 TSE sequences for hippocampal segmentation. The DRB method improves efficiency while providing clinically reliable results, and the proposed 95% confidence interval can aid in more objective assessments of hippocampal pathology.