Application of a semiautomatic classifier for modic and disk hernia changes in magnetic resonance

OBJECTIVE: Early detection of degenerative changes in lumbar intervertebral disc by magnetic resonance imaging in a semiautomatic classifier for prevention of degenerative disease. METHOD: MRIs were selected with a diagnosis of degenerative disc disease or back pain from January to May 2014, with a sample of 23 patients and a total of 170 disks evaluated by sagittal T2 MRI image, first evaluated by a specialist physician in training and them were introduced into the software, being the results compared. RESULTS: One hundred and fifteen discs were evaluated by a programmed semiautomatic classifier to identify MODIC changes and hernia, which produced results "normal or MODIC" and "normal or abnormal", respectively. With a total of 230 readings, of which 141 were correct, 84 were reading errors and 10 readings were undiagnosed, the semiautomatic classifier is a useful tool for early diagnosis or established disease and is easy to apply because of the speed and ease of use; however, at this early stage of development, software is inferior to clinical observations and the results were from around 65% to 60% certainty for MODIC rating and 61% to 58% for disc herniation, compared with clinical evaluations. CONCLUSION: The comparative results between the two doctors were 94 consistent results and only 21 errors, which represents 81% certainty.

研究目标：通过半自动分类器结合磁共振成像（Magnetic Resonance Imaging, MRI），早期检测腰椎间盘退行性病变，以实现退行性疾病的预防。
研究方法：选取2014年1月至5月期间确诊为椎间盘退行性疾病或腰痛患者的影像学数据，共纳入23例受试者，总计170个椎间盘通过矢状位T2加权MRI图像完成评估。所有图像先由受训专科医师进行初步阅片，随后将图像导入分析软件，并将两类结果进行对比。
研究结果：通过预设的半自动分类器对115个椎间盘进行评估，以识别MODIC改变与椎间盘突出，分类器分别输出"正常或MODIC改变"、"正常或异常"两类结果。本次总计完成230次阅片，其中141次结果正确，84次为阅片误差，10次未获得明确诊断。该半自动分类器可用于早期或已确诊疾病的辅助诊断，且操作简便、分析速度快，具备较高应用价值；但在当前开发初期阶段，该软件性能仍逊于临床医师阅片：与临床评估结果相比，MODIC分级的准确率约为60%~65%，椎间盘突出识别准确率约为58%~61%。
研究结论：两位医师的阅片对比结果显示，94次结果一致，仅21次出现误差，一致性准确率达81%。