Overview
Low back pain (LBP) is worldwide responsible for more years lived with disability than any other health condition. In the Netherlands, approximately 44% of the population experiences at least one episode of LBP in their lifetime, with one in five reporting persistent back pain lasting longer than three months (chronic low back pain, CLBP). CLBP often results in substantial limitations in functional activities and is responsible for high healthcare and socioeconomic costs. In the vast majority of patients with CLBP (85-90%), the aetiology is unknown and for medical specialists it is challenging to identify patients who would benefit from surgical or non-surgical interventions.
The Nijmegen Decision Tool for CLBP (NDT-CLBP) is a pre-diagnostic decision support tool that matches patients based on a questionnaires to the treatment that they are most likely to benefit from. Patients are referred to either spinal surgeon consultation or non-surgical consultation. In this project the NDT-CLBP will be further developed into a two-phased decision support tool following the patient journey: phase 1 decision support for consultation (currently in use) and phase 2 decision support for referral to a specific treatment based on the clinical diagnostic phase (NDT-CLBP 2.0). In the diagnostic phase, many patients with CLBP receive a lumbar spine MRI scan to detect degenerative changes of the spine. The goal of this project is to develop an AI-based image analysis algorithms that enable detailed quantitative routine analysis of these MRI scans.
MRI image features with an evident relation to low back pain: a narrative review
To identify which MR image features are related to CLBP, a narrative review was written. This narrative review provides an overview of all possible relevant features visible on MRI images and determines their relation to LBP. We conducted a separate literature search per image feature. The various relations between MRI features and their associated pain mechanisms were evaluated to provide a list of features that are related to LBP. All searches combined generated a total of 4472 hits, of which 31 articles were included. Features were divided into five different categories: 'discogenic', 'neuropathic', 'osseous', 'facetogenic', and 'paraspinal', and discussed separately. Our research suggests that type I Modic changes, disc degeneration, endplate defects, disc herniation, spinal canal stenosis, nerve compression, and muscle fat infiltration have the highest probability of being related to LBP. These can be used to improve clinical decision-making for patients with LBP based on MRI. The full paper can be found here.
Lumbar spine segmentation in MR images: A dataset and deep learning algorithm
The first step in creating an algorithm for lumbar spine analysis is getting an automatic segmentation of the spine. In our first project, we aim to create a convolutional neural network that automatically segments vertebrae and intervertebral discs on MRI scans. Anonymized clinical MRI scans from the RadboudUMC are manually annotated, which will be used to train the network. The results of an early version of this network were published as a conference paper, winning the prize for the best poster presentation. The current network now also segments the spinal canal and is trained on a much larger dataset. This dataset will be made publicly available alongside a segmentation challenge (spider.grand-challenge.org), which is currently being set up. The segmentation algorithm can be found here.
An automatic method to measure Cobb angles in MRI of patients with adult spinal deformity using artificial intelligence
This study aimed to evaluate a novel automatic method for measuring coronal Cobb angles on sagittal lumbar MRI scans in patients with adult spinal deformity (ASD). Using the segmentation algorithm described earlier, we automatically segmented vertebrae and intervertebral discs. We then fitted 3D planes through the upper and lower half of each individual intervertebral disc. All possible angles between these planes were measured, and the largest AP angle was extracted, which served as an estimate of the Cobb angle. We compared our automatic measurements to manual measurements by three experienced readers and found that our algorithm had similar accuracy to that of human readers. Our findings suggest that this automated method can reduce workload and subjectivity in monitoring the curve progression of ASD.
