基于深度残差网络的脊柱疾病分类预测

Classification and Prediction of Spine Diseases Based on Deep Residual Network

脊柱退化性疾病,作为脊柱外科中最常见的疾病脊椎疾病,其发病正呈现年轻化的趋势。磁共振成像作为一种非侵入式检查手段,凭借对软组织成像好、无辐射、对肌肉骨骼疾病的特异性和敏感度较高等优点,在临床上被用于脊椎疾病的诊断。该文提出了一种基于深度残差网络的脊椎病核磁共振图像人工智能分类模型,能够帮助医生实现早期的脊柱退化性疾病筛查,帮助患者尽早得到正确有效的治疗。据实验结果表明,该模型不仅可以获得比传统神经网络深度学习算法更高的脊椎疾病识别率,还可以比传统神经网络深度学习算法提高35%到85%的计算效率并节省70%以上的内存占用。这一点使得该算法能够适应于资源有限的移动终端和对延迟要求比较高的医疗场景。

Spinal degeneration disease, the most common disease in spinal surgery, has become a problem for the old and even the young people. Magnetic resonance imaging, as a non-invasive examination method, is clinically used in the diagnosis of spinal diseases due to its advantages of good soft tissue imaging, no radiation, high specificity and sensitivity to musculoskeletal diseases. Therefore, we propose a deep residual network based artificial intelligence classification model for MRI images of spinal diseases, which can help doctors to realize early screening of spinal degenerative diseases and help patients to get correct and effective treatment as soon as possible. According to the experimental results, the proposed model can not only obtain a higher recognition rate of spinal diseases than the traditional neural network deep learning algorithm, but also improve the computational efficiency by 35% to 85% and save more than 70% of the memory. This enables the algorithm to be suitable for mobile terminals with limited resources and medical scenarios with high latency requirements.