Deep Learning for Multi-Tissue Segmentation and Fully Automatic Personalized Biomechanical Models from BACPAC Clinical Lumbar Spine MRI

Abstract

Structured Abstract Study Design In vivo retrospective study of fully automatic quantitative imaging feature extraction from clinically acquired lumbar spine magnetic resonance imaging (MRI). Objective To demonstrate the feasibility of substituting automatic for human-demarcated segmentation of major anatomical structures in clinical lumbar spine MRI to generate quantitative image-based features and biomechanical models. Setting Previous studies have demonstrated the viability of automatic segmentation applied to medical images; however, the feasibility of these networks to segment clinically acquired images has not yet been demonstrated, as they largely rely on specialized sequences or strict quality of imaging data to achieve good performance. Methods Convolutional neural networks were trained to demarcate vertebral bodies, intervertebral disc, and paraspinous muscles from sagittal and axial T1-weighted MRIs. Intervertebral disc height, muscle cross sectional area, and subject-specific musculoskeletal models of tissue loading in the lumbar spine were then computed from these segmentations and compared against those computed from human-demarcated masks. Results Segmentation masks, as well as the morphological metrics and biomechanical models computed from those masks, were highly similar between human- and computer-generated methods. Segmentations were similar with Dice Similarity Coefficients 0.77 or greater across networks, morphological metrics and biomechanical models were similar with Pearson R correlation coefficients 0.69 or greater when significant. Conclusions This study demonstrates the feasibility of substituting computer-generated for human-generated segmentations of major anatomical structures in lumbar spine MRI to compute quantitative image-based morphological metrics and subject-specific musculoskeletal models of tissue loading quickly, efficiently, and at scale without interrupting routine clinical care.

MRI‐based measurement of in vivo disc mechanics in a young population due to flexion, extension, and diurnal loading

“…Image registration quality was evaluated for all 240 registrations using Dice similarity coefficient (where a Dice of 0 means there was zero overlap of the objects, while 1 means perfect overlap of the objects), where the loaded image's manual disc segmentation was transformed to the Reference image via the calculated registration and then compared to the Reference image's manual disc segmentation. Dice coefficient was high for this study, indicating strong registrations (0.92 ± 0.03), comparable to the best of prior literature 55,56 …”

“…high for this study, indicating strong registrations (0.92 ± 0.03), comparable to the best of prior literature. 55,56 The anatomic disc coordinate system was defined for each disc with its left-right axis the same as the left-right patient axis from the MRI scanner. Maintaining both this constraint and orthogonality, the A-P and axial (superior-inferior) axes were set to align as closely as possible with the second and third principal axes of the segmented disc volume determined via principal component analysis.…”

“…Dice coefficient was high for this study, indicating strong registrations (0.92 ± 0.03), comparable to the best of prior literature. 55 , 56 …”

Artificial Intelligence to Preoperatively Predict Proximal Junction Kyphosis Following Adult Spinal Deformity Surgery: Soft Tissue Imaging May Be Necessary for Accurate Models

“…Thus, the accuracy of model 3 in conjunction with the Grad-CAM analysis, provides evidence that muscularity and fatty degeneration as observable on MRI may offer the highest accuracy to predict PJK. To better quantify the exact tissue composition of these predictive MRI features, future work could leverage existing techniques to segment tissue types automatically 43 …”

“…To better quantify the exact tissue composition of these predictive MRI features, future work could leverage existing techniques to segment tissue types automatically. 43 Of greatest clinical interest is how to alter surgical decision-making, patient counseling, and postoperative management of patients predicted to develop PJK. Perhaps, a future model could be developed that individually incorporates all possible combinations of surgical decisions (e.g.…”