Motivation¶

Low back pain (LBP) is a significant cause of disability and a major societal healthcare issue. One of the commonly used diagnostic and treatment decision-making tools for LBP is magnetic resonance imaging (MRI) of the lumbar spine. Over the past few decades, the use of MRI for patients with LBP has increased substantially. Automatic image analysis has the potential to alleviate the increased workload for radiologists and spinal surgeons and improve the diagnostic value of MRI by enabling more objective and quantitative image interpretation. However, to effectively evaluate complex multifactorial disorders such as LBP, automatic analysis must comprehend multiple anatomical elements of the spine, including the vertebrae, the intervertebral discs (IVDs), and the spinal canal. Therefore, a robust automatic algorithm for segmenting these structures is essential.

To develop accurate and reliable AI algorithms for lumbar spine MRI segmentation, large and diverse datasets with reference segmentations are needed. However, currently available datasets for lumbar spine MRI segmentation are either limited in size, only segment one or two anatomical structures, or are not publicly available. To address this gap, we introduce a large multi-center lumbar spine MR dataset with reference segmentations of vertebrae, intervertebral discs (IVDs), and spinal canal, which we make publicly available to serve as training data for AI algorithms.

To foster the development of new and improved lumbar spine MRI segmentation algorithms, we also introduce a continuous segmentation challenge. Participants are invited to submit their AI models for evaluation on a hidden test set, which was held back from the publicly available training data. The challenge provides an opportunity for algorithm developers to benchmark their models against other state-of-the-art algorithms and to contribute to the advancement of lumbar spine MRI segmentation research.

Task¶

The Lumbar SPIDER Challenge focuses on the segmentation of three anatomical structures in lumbar spine MRI: vertebrae, intervertebral discs (IVDs), and spinal canal. The segmentation task requires participants to produce separate masks for each vertebra, IVD, and the spinal canal in the lumbar spine MRI volume. The numbering of the vertebrae and IVDs is not specific and may vary across different cases.

The paper on the Challenge, dataset, and the baseline algorithms can be found here: https://www.nature.com/articles/s41597-024-03090-w