CLASSICAL TECHNIQUES FOR SPINAL LESION DETECTION: A BASELINE STUDY AND FUTURE OUTLOOK

Abstract

This study presents a comprehensive pipeline for spinal cord lesion segmentation in magnetic resonance images using classical image processing techniques implemented in Python. The dataset comprises one real spinal magnetic resonance image and five synthetically generated cases to ensure robustness and diversity. Our workflow, comprising grayscale conversion, 8‑bit normalization, Gaussian blurring for noise reduction, Canny edge detection, and threshold‑based segmentation, was quantitatively evaluated using the Dice similarity coefficient and Intersection over Unionmetrics. For the real case, we obtained a Dice score of 0.78 and an Intersection over Unionmetrics of 0.65; across the synthetic cases, the average Dice was 0.82 and the Intersection over Unionmetrics was 0.70. These results demonstrate that classical image processing methods can reliably delineate lesion regions with high computational efficiency and interpretability, making them suitable for preliminary analysis and label generation in resource‑constrained clinical environments. Future work will focus on expanding the real‑patient dataset, implementing adaptive thresholding, and integrating deep learning–based enhancements to improve generalizability.

Keywords

• spinal cord lesion segmentation
• magnetic resonance imaging
• Dice similarity coefficient

References

1. [1] Zhang, Y., Yang, G., & Zhang, J. (2019). Brightness-based hemorrhage segmentation in CT scans using classical image processing techniques. Journal of Medical Imaging and Health Informatics, 9(3), 512–519.
2. [2] Gonca, A. (2022). Meme kanserinde makine öğrenmesi ile erken tanı sisteminin geliştirilmesi (Yüksek Lisans Tezi, İzmir Demokrasi Üniversitesi). Türkiye Ulusal Tez Merkezi.
3. [3] Çakmakcı, H. (2021). Görüntü işleme teknolojisi üzerine (Yüksek Lisans Tezi, Giresun Üniversitesi, Fen Bilimleri Enstitüsü). Türkiye Ulusal Tez Merkezi.
4. [4] Yüzbaşı, C. (2020). 4-D matrisler üzerinde cebirsel işlemler (Yüksek Lisans Tezi, Gaziantep Üniversitesi, Fen Bilimleri Enstitüsü, Matematik Anabilim Dalı). Türkiye Ulusal Tez Merkezi.
5. [5] Taher, S. M., & Ghanim, M. (2023). Applied improved Canny edge detection for diagnosis medical images of human brain tumors. Al Mustansiriyah Journal of Science, 34(4), 66–74.
6. [6] Gonzalez, R. C., & Woods, R. E. (2018). Digital image processing (4th ed.). Pearson Education.
7. [7] Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, 9351, 234–241.
8. [8] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770–778.

9. [9] Wang, X., Girshick, R., Gupta, A., & He, K. (2018). Non-local neural networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 7794–7803. https://doi.org/10.1109/CVPR.2018.00813