Background and Objective: This article describes a fully automatic system for classifying various spinal degenerative phenotypes namely Modic changes, endplate defects and focal changes which are associated with lower back pain. These are obtained from T1/T2 Magnetic Resonance Imaging (MRI) scans. Lower back pain is a predominantly occurring ailment, which is prone to have various roots including the anatomical and pathophysciological aspects. Clinicians and radiologist use MRI to assess and evaluate the extent of damage, cause, and to decide on the future course of treatment. In large healthcare systems, to circumvent the manual reading of various image slices, we describe a system to automate the classification of various vertebral degeneracies that cause lower back pain. Methods: We implement a combination of feature extraction, image analysis based on geometry and classification using machine learning techniques for identifying vertebral degeneracies. Image features like local binary pattern, Hu’s moments and gray level co-occurrence matrix (GLCM) based features are extracted to identify Modic changes, endplate defects, and presence of any focal changes. A combination of feature set is used for describing the extent of Modic change on the end plate. Feature sensitivity studies towards efficient classification is presented. A STIR based acute/chronic classification is also attempted in the current work. Results: The implemented method is tested and validated over a dataset containing 100 patients. The proposed framework for detecting the extent of Modic change achieves an accuracy of 85.91%. From the feature sensitivity analysis, it is revealed that entropy based measure obtained from gray level co-occurrence matrix alone is sufficient for detection of focal changes. The classification performance for detecting endplate defect is highly sensitive to the first 2 Hu’s moments. Conclusion: A novel approach to identify the allied vertebral degenerations and extent of Modic changes in vertebrae by exploiting image features and classification through machine learning is proposed. This shall assist radiologists in detecting abnormalities and in treatment planning.

背景和目的：本文描述了一种全自动系统，用于分类各种脊柱退行性表型，即与下背痛相关的莫迪奇变化、终板缺损和局灶性变化。这些是从 T1/T2 磁共振成像 (MRI) 扫描中获得的。腰痛是一种主要发生的疾病，它容易有各种根源，包括解剖学和病理生理学方面。临床医生和放射科医生使用 MRI 来评估和评估损伤程度、原因，并决定未来的治疗方案。在大型医疗保健系统中，为了避免手动读取各种图像切片，我们描述了一种系统来自动分类导致腰痛的各种椎体退化。方法：我们实现了特征提取的组合，基于几何和分类的图像分析，使用机器学习技术识别椎体退化。提取图像特征，如局部二值模式、胡氏矩和基于灰度共生矩阵 (GLCM) 的特征，以识别 Modic 变化、终板缺陷和任何焦点变化的存在。特征集的组合用于描述端板上Modic变化的程度。提出了对有效分类的特征敏感性研究。在当前的工作中也尝试了基于 STIR 的急性/慢性分类。结果：实施的方法在包含 100 名患者的数据集上进行了测试和验证。用于检测 Modic 变化程度的拟议框架达到了 85.91% 的准确度。从特征敏感性分析来看，结果表明，仅从灰度共​​生矩阵获得的基于熵的测量就足以检测焦点变化。检测端板缺陷的分类性能对前 2 Hu 矩高度敏感。结论：提出了一种通过机器学习利用图像特征和分类来识别椎骨相关椎体退变和 Modic 变化程度的新方法。这将有助于放射科医生检测异常情况和制定治疗计划。提出了一种通过机器学习利用图像特征和分类来识别椎骨相关椎体退化和Modic变化程度的新方法。这将有助于放射科医生检测异常情况和制定治疗计划。提出了一种通过机器学习利用图像特征和分类来识别椎骨相关椎体退化和Modic变化程度的新方法。这将有助于放射科医生检测异常情况和制定治疗计划。