Abnormality Detection in Spinal Cord MRI Using K-Means Clustering

Abstract

Early and accurate detection of spinal cord abnormalities is crucial for preventing severe neurological deficits and enabling timely medical interventions. Magnetic Resonance Imaging (MRI) provides detailed visualization of spinal anatomy, yet manual inspection is time-consuming and prone to subjective errors. This study presents an automated approach for abnormality detection in spinal cord MRI images using unsupervised clustering techniques. Histogram-based features are extracted from both DICOM and JPEG formats, followed by preprocessing and standardization to construct a robust feature space. Multiple clustering algorithms, including K-Means, BIRCH, and DBSCAN, are applied to identify abnormal patterns, with evaluation conducted using Silhouette, Davies-Bouldin, and Calinski-Harabasz indices. Results indicate that K-Means outperforms the other methods in terms of clustering cohesion and separability, with lower computational time and higher interpretability. Anomaly scores and Z-scores are computed to quantify deviations from normal patterns, enabling severity-based assessment. Additionally, a Flask-based front-end interface integrated with SQLite3 facilitates user authentication and interactive image upload, allowing clinicians to visualize clustering results and abnormality severity in real-time. This approach demonstrates the potential of combining histogram features with unsupervised clustering for efficient and scalable spinal cord abnormality detection.