基于MRI鉴别诊断结核性与化脓性脊柱炎的研究进展与模型分析

Research Progress and Model Analysis of Differential Diagnosis between Tuberculous and Pyogenic Spondylitis Based on MRI

摘要: 结核性脊柱炎和化脓性脊柱炎的临床表现比较类似，但是它们的治疗策略以及预后差别很大，所以，临床医生要尽早而且精确地鉴别诊断这两种病，这是一项关键性的难点任务。磁共振成像有着很强的软组织分辨能力，所以它成了考量脊柱炎性病变时最为理想的一种医学成像技术，这篇综述希望全面整理依靠MRI影像学特点来形成结核性脊柱炎和化脓性脊柱炎区别诊断模型的相关研究动态。近些年来，伴随人工智能技术不断向前迈进，依托MRI的影像组学以及深度学习模型渐渐成为研究热点，这些模型可以高效率地获取并剖析人眼不易察觉到的深层次影像特征，进而塑造起客观又具量化性质的判断工具，相关研究显示，这种先进的模型在辨别这两种病症的时候，其准确率、灵敏度和特异度均表现出大幅的优势，甚至超过了放射科医生日常的判别水准，由此得出的结论表明，把传统影响特征同人工智能技术结合在一起的区分诊断模型属于做到精准区分的重要发展走向，日后的研究应当着眼于大规模、多中心的外部验证工作，促使该模型深入到临床指标当中去，从而加快其朝着临床应用方向前进的步伐。

Abstract: The clinical manifestations of tuberculous spondylitis and pyogenic spondylitis are quite similar; however, their treatment strategies and prognoses differ significantly. Therefore, it is a crucial and challenging task for clinicians to achieve early and accurate differential diagnosis of these two conditions. Magnetic resonance imaging (MRI), with its strong soft tissue resolution, has become the ideal medical imaging technique for evaluating spinal inflammatory lesions. This review aims to comprehensively summarize the current research landscape regarding differential diagnosis models for tuberculous spondylitis and pyogenic spondylitis based on MRI characteristics. In recent years, with the continuous advancement of artificial intelligence (AI) technology, MRI-based radiomics and deep learning models have gradually become research hotspots. These models can efficiently extract and analyze subtle, deep-level imaging features that are difficult to discern with the naked eye, thereby establishing objective and quantitative diagnostic tools. Relevant studies have shown that such advanced models demonstrate significant advantages in distinguishing between these two diseases in terms of accuracy, sensitivity, and specificity, even surpassing the diagnostic performance of radiologists in routine practice. Consequently, the findings indicate that integrating traditional imaging features with AI technology to create differential diagnosis models represents a major development direction for achieving precise differentiation. Future research should focus on large-scale, multi-center external validation to integrate these models into clinical practice guidelines, thereby accelerating their translation into clinical applications.