Yapay Zeka Hızlı Bakış & Çıkarımlar
AI InsightFuture work will focus on expanding the real‑patient dataset, implementing adaptive thresholding, and integrating deep learning–based enhancements to improve generalizability.
📌 3 Maddede Bu Makale
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.