Bel Omurgası Dejeneratif Hastalıklarının Derin Öğrenme Teknikleri ile Sınıflandırılması

Öz

Bu çalışma, bel omurgası dejeneratif hastalıklarının sınıflandırılması için derin öğrenme (DL) tekniklerinin kullanımını incelemektedir. Özellikle çalışmada spinal kanal stenozunun tespitinde kullanılan Manyetik Rezonans (MR) görüntüleri değerlendirilmiştir. Derin öğrenme modellerinin, radyolojik görüntüler üzerinde otomatik analiz yeteneği sayesinde tanı süreçlerini hızlandırma potansiyeli ortaya konulmuştur. Çalışmada, farklı derin öğrenme modelleri kullanılmış ancak en düşük kayıp değeri EfficientNetV2-Large mimarisi ile elde edilmiştir. Gelişmiş veri artırma teknikleri, özellikle nadir görülen vakalar için hedefli yaklaşımlar ve yüksek çözünürlüklü (512x512) görüntülerle çalışılması, modelin performansını önemli ölçüde artırmıştır. Mimari güncellemeler ve veri işleme stratejileri sonucunda, test log-loss değeri en düşük 0.69 seviyesine düşürülmüştür. Ayrıca, farklı modellerin tahminlerinin ensemble learning (topluluk öğrenmesi) çerçevesinde soft voting yöntemiyle bir araya getirilmesiyle elde edilen sonuçlar da sunulmuştur. Bu yaklaşım, public test veri setinde 0.604510 gibi düşük bir log-loss değeri elde etmiştir. Sonuçlar, modelin klinik açıdan kritik olan "severe" vakaları ayırt etme yeteneğini ve genişletilmiş sınıf yapısında dahi genelleme gücünü koruduğunu göstermektedir.

Anahtar Kelimeler

• Dejeneratif disk hastalığı
• tıbbi görüntüleme
• Data Augmentation
• CNN
• Ensemble Learning.

Classification of Lumbar Spine Degenerative Diseases Using Deep Learning Techniques

Abstract

This study investigates the use of deep learning (DL) techniques for the classification of lumbar spine degenerative diseases. In particular, Magnetic Resonance (MR) images used for the detection of spinal canal stenosis are evaluated. The potential of deep learning models to accelerate diagnostic processes through their capability for automatic analysis of radiological images is demonstrated. Various deep learning models were employed in the study; however, the lowest loss value was achieved with the EfficientNetV2-Large architecture. Advanced data augmentation techniques, especially targeted approaches for rare cases, and the use of high-resolution (512x512) images significantly improved the model's performance. As a result of architectural updates and data processing strategies, the test log-loss value was reduced to as low as 0.69. Additionally, the results obtained by combining the predictions of different models through ensemble learning with a soft voting method are also presented. This approach yielded a low log-loss value of 0.604510 on the public test dataset. The results demonstrate that the model is capable of distinguishing clinically critical "severe" cases and maintains its generalization ability even in an expanded class structure.

Keywords

• Degenerative disc disease
• medical imaging
• Data Augmentation
• CNN
• Ensemble Learning.

References

1. [1] K. Hoffeld et al., "Patient-related risk factors and lifestyle factors for lumbar degenerative disc disease: a systematic review," Neurochirurgie, vol. 69, no. 5, p. 101482, 2023.
2. [2] W. Liawrungrueang, J.-B. Park, W. Cholamjiak, P. Sarasombath, and K. D. Riew, "Artificial intelligence-assisted MRI diagnosis in lumbar degenerative disc disease: a systematic review," Global Spine Journal, vol. 15, no. 2, pp. 1405-1418, 2025.
3. [3] L. Scarcia et al., "Degenerative disc disease of the spine: from anatomy to pathophysiology and radiological appearance, with morphological and functional considerations," Journal of Personalized Medicine, vol. 12, no. 11, p. 1810, 2022.
4. [4] R. U. Din, X. Cheng, and H. Yang, "Diagnostic role of magnetic resonance imaging in low back pain caused by vertebral endplate degeneration," Journal of Magnetic Resonance Imaging, vol. 55, no. 3, pp. 755-771, 2022.
5. [5] M. Hussain, D. Koundal, and J. Manhas, "Deep learning-based diagnosis of disc degenerative diseases using MRI: a comprehensive review," Computers and Electrical Engineering, vol. 105, p. 108524, 2023.
6. [6] H. B. Viral, "Advanced Classification of Lumbar Spine Degenerative Disorders Using Spine-CNN Attenuation Model," International Journal of Intelligent Systems and Applications in Engineering, vol. 12, no. 3, pp. 3687 – 3694, 03/24 2024. [Online]. Available: https://ijisae.org/index.php/IJISAE/article/view/6045.
7. [7] Z. Wang, P. Xiao, and H. Tan, "Spinal magnetic resonance image segmentation based on U-net," Journal of Radiation Research and Applied Sciences, vol. 16, no. 3, p. 100627, 2023.
8. [8] R. Pal, P. Saha, S. Ghoshal, A. Chakrabarti, and S. Sur-Kolay, "Panoptic Segmentation and Labelling of Lumbar Spine Vertebrae using Modified Attention Unet," arXiv preprint arXiv:2404.18291, 2024.

9. [9] F. Milletari, N. Navab, and S.-A. Ahmadi, "V-net: Fully convolutional neural networks for volumetric medical image segmentation," in 2016 fourth international conference on 3D vision (3DV), 2016: Ieee, pp. 565-571.
10. [10] J. T. Tyler Richards, Robyn Ball, Errol Colak, Adam Flanders, Felipe Kitamura, John Mongan, Luciano Prevedello, and Maryam Vazirabad, "RSNA 2024 Lumbar Spine Degenerative Classification," Kaggle, 2024. [Online]. Available: https://kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification.
11. [11] D. C. Preston, "Normal mid sagittal MRI scans of the lumbar spine (T1 and T2 weighted images)," Case Western Reserve University., 2006. [Online]. Available: https://case.edu/med/neurology/NR/MRI_Spine/NormalSagittalMRI.htm.
12. [12] D. C. Preston, "Axial MRI of the lumbar spine – T2 weighted image at the L4 level [Image]. Case Western Reserve University," 2006. [Online]. Available: https://case.edu/med/neurology/NR/MRI_Spine/nl%20ls%20anatomy.htm.
13. [13] Radiopaedia, "Sagittal STIR MRI of the lumbar spine [Magnetic resonance imaging]. Radiopaedia.org.," 2023. [Online]. Available: https://radiopaedia.org/images/53402207.