Multi-Pathology Segmentation of the Lumbar Spine

• Claudio Leite UFSCar
• Jurandy Almeida UFSCar

Resumo

The diagnosis of spinal pathologies is complex due to the frequent overlap of multiple diseases in the same anatomical location, a scenario that current segmentation or classification methods do not efficiently address. This work presents an empirical study on the segmentation of multiple overlapping pathologies, proposing and systematically comparing three strategies: (i) a baseline binary class approach using independent models; (ii) a multi-class approach mapping disease combinations to unique labels; and (iii) a multi-label approach using parallel channels to explicitly model co-occurrence.We evaluated over 300 training and inference pipelines, combining five neural network architectures and three loss functions. Our preliminary results show that the multi-label strategy significantly outperforms the other approaches in both accuracy and computational efficiency, establishing a promising direction for developing robust, scalable diagnostic tools.

Referências

G. McNicoll, ”World Population Ageing 1950-2050,” Popul. Dev. Rev., pp. 814-816, 2002.

M.G. Fehlings et al., ”The aging of the global population: the changing epidemiology of disease and spinal disorders,” Neurosurgery, pp. S1-S5, 2015.

J. W. van der Graaf et al., “Lumbar spine segmentation in MR images: a dataset and a public benchmark,” Sci. Data, vol. 11, no. 264, 2024.

İ. Altun et al., ”LSS-UNET: Lumbar spinal stenosis semantic segmentation using deep learning,” Multimed. Tools Appl., 82:41287-41305, 2023.

J. Qian et al., ”Lumbar disc herniation diagnosis using deep learning on MRI,” J. Radiat. Res. Appl. Sci., 17(3):100988, 2024.

R. Windsor et al., ”SpineNetV2: automated detection, labelling and radiological grading of clinical MR scans,” arXiv:2205.01683, 2022.

A. Hatamizadeh et al., ”UNETR: Transformers for 3D medical image segmentation,” in WACV, pp. 1748-1758, 2022.

E. Kerfoot et al., ”Left-ventricle quantification using residual U-Net,” in STACOM, LNCS 11395, pp. 371-380, 2019.

F. Milletari et al., ”V-Net: Fully convolutional neural networks for volumetric medical image segmentation,” in 3DV, pp. 565-571, 2016.

A. Hatamizadeh et al., ”Swin UNETR: Swin Transformers for semantic segmentation of brain tumors in MRI images,” in BrainLes, LNCS 12962, pp. 272-284, 2022.

K. He et al., ”Deep residual learning for image recognition,” in CVPR, pp. 770-778, 2016.

T.-Y. Lin et al., ”Focal loss for dense object detection,” in ICCV, pp. 2980-2988, 2017.

C. H. Sudre et al., ”Generalised Dice Overlap as a deep learning loss function for highly unbalanced segmentations,” in DLMIA, MICCAI Wkshp., pp. 240-248, 2017.

W. Mbarki et al., ”A novel method based on deep learning for herniated lumbar disc segmentation,” in IC ASET, pp. 394-399, 2020.

J. Qian et al., ”Lumbar disc herniation diagnosis using deep learning on MRI,” J. Radiat. Res. Appl. Sci., 17(3):100988, 2024.

Q. Pan et al., ”Automatically diagnosing disk bulge and disk herniation...: Method development study,” JMIR Med. Inform., 9(5):e14755, 2021.

Y. Chen et al., ”Deep learning-based computer-aided diagnostic system for lumbar degenerative diseases classification using MRI,” Biomed. Signal Process. Control, 109:108002, 2025.

Y. Wang et al., ”Deep learning-driven diagnosis of multi-type vertebra diseases based on computed tomography images,” Quant. Imaging Med. Surg., 14(1):800, 2023.

M.-L. Zhang and Z.-H. Zhou, ”A review on multi-label learning algorithms,” IEEE Trans. Knowl. Data Eng., 26(8):1819-1837, 2013.