Automatic segmentation of the spinal cord and intramedullary multiple sclerosis lesions with convolutional neural networks (1805.06349v2)

Abstract: The spinal cord is frequently affected by atrophy and/or lesions in multiple sclerosis (MS) patients. Segmentation of the spinal cord and lesions from MRI data provides measures of damage, which are key criteria for the diagnosis, prognosis, and longitudinal monitoring in MS. Automating this operation eliminates inter-rater variability and increases the efficiency of large-throughput analysis pipelines. Robust and reliable segmentation across multi-site spinal cord data is challenging because of the large variability related to acquisition parameters and image artifacts. The goal of this study was to develop a fully-automatic framework, robust to variability in both image parameters and clinical condition, for segmentation of the spinal cord and intramedullary MS lesions from conventional MRI data. Scans of 1,042 subjects (459 healthy controls, 471 MS patients, and 112 with other spinal pathologies) were included in this multi-site study (n=30). Data spanned three contrasts (T1-, T2-, and T2*-weighted) for a total of 1,943 volumes. The proposed cord and lesion automatic segmentation approach is based on a sequence of two Convolutional Neural Networks (CNNs). To deal with the very small proportion of spinal cord and/or lesion voxels compared to the rest of the volume, a first CNN with 2D dilated convolutions detects the spinal cord centerline, followed by a second CNN with 3D convolutions that segments the spinal cord and/or lesions. When compared against manual segmentation, our CNN-based approach showed a median Dice of 95% vs. 88% for PropSeg, a state-of-the-art spinal cord segmentation method. Regarding lesion segmentation on MS data, our framework provided a Dice of 60%, a relative volume difference of -15%, and a lesion-wise detection sensitivity and precision of 83% and 77%, respectively. The proposed framework is open-source and readily available in the Spinal Cord Toolbox.

• The paper presents a novel two-stage CNN method that segments the spinal cord and MS lesions from multi-site MRI data.
• It achieves a median Dice coefficient of 95% for the cord and 60% for lesions, outperforming traditional segmentation methods.
• The framework enhances clinical reliability by reducing manual input and inter-rater variability in long-term MS monitoring.

Automatic Segmentation of the Spinal Cord and Intramedullary Multiple Sclerosis Lesions with Convolutional Neural Networks

The paper presents a novel approach for the automatic segmentation of spinal cord and intramedullary multiple sclerosis (MS) lesions using convolutional neural networks (CNNs). The paper leverages large datasets from multiple sites to train and evaluate the robustness of a deep learning framework, ultimately addressing the challenges in multi-site MRI segmentation tasks posed by variability in acquisition parameters, anatomical heterogeneity, and artifact presence.

In efforts to segment the spinal cord and MS lesions from conventional MRI data, the authors propose a framework consisting of two CNNs. The first CNN employs 2D dilated convolutions to detect the spinal cord centerline. This is followed by a second CNN employing 3D convolutions to segment the spinal cord and/or lesions surrounding the inferred centerline. This architecture effectively tackles the image variability and allows for the segmentation process to be robust over different MRI contrasts and resolutions.

The dataset used for training includes scans from 1,042 subjects—comprising MS patients, healthy controls, and individuals with other spinal pathologies. Training and evaluation cover three MRI contrasts: T1-, T2-, and T2'-weighted, ensuring the model can accommodate various acquisition parameters. The CNNs were trained using Dice loss to mitigate class imbalance issues due to the sparsity of lesion voxels, demonstrating significantly improved performance over traditional methods.

Numerically, the CNN-based framework achieved a median Dice coefficient of 95% for spinal cord segmentation and 60% for MS lesion segmentation. The spinal cord segmentation outperformed PropSeg, a prevalent spinal cord segmentation method, which reported a lower median Dice of 88%. Notably, the lesion segmentation revealed a median relative volume difference of -15%. Voxel-wise sensitivity and precision for lesion detection showed values of 83% and 77%, respectively.

The implications of these findings are multi-fold; the automated method provides a reliable, efficient tool for spinal cord and MS lesion segmentation across diverse clinical datasets, reducing manual input and minimizing inter-rater variability. This framework is valuable for longitudinal monitoring of MS progression and development of atrophy, offering consistency crucial for clinical applications. Furthermore, the paper underscores the potential of CNNs to adapt to varying conditions, laying groundwork for future applications in AI-driven medical imaging where data inconsistency is prevalent.

Looking forward, the research explores the inclusion of additional MRI contrasts beyond the T1, T2, and T2', such as diffusion-weighted and echo-planar imaging, to further augment the segmentation capabilities of the model. Additionally, improvements in specificity and reduction of false positives in lesion segmentation could be realized through expanded datasets and enhanced preprocessing techniques.

Ultimately, this paper illustrates the utility and efficacy of deep learning in handling complex medical imaging challenges, demonstrating significant advancements in spinal cord and MS lesion segmentation. The methods presented reflect significant strides in integrating AI to streamline clinical analysis in neurology and potentially other medical domains.