Endocrinol Metab (Seoul). 2022;37(4):674-683.

Development of a Spine X-Ray-Based Fracture Prediction Model Using a Deep Learning Algorithm.

, , , , , ,

제1저자 공성혜, 이재원 · 교신저자 김정희

PMID 35927066 PMC9449110 DOI 10.3803/EnM.2022.1461 IF 5.1 내분비대사내과

소속 (4)

Department of Internal Medicine, Seoul National University Bundang Hospital, Seongnam, Korea.
Department of Internal Medicine, Seoul National University College of Medicine, Seoul, Korea.
VUNO Inc., Seoul, Korea.
Department of Internal Medicine, Seoul National University Hospital, Seoul, Korea.

Abstract

BACKGRUOUND Since image-based fracture prediction models using deep learning are lacking, we aimed to develop an X-ray-based fracture prediction model using deep learning with longitudinal data.

METHODS This study included 1,595 participants aged 50 to 75 years with at least two lumbosacral radiographs without baseline fractures from 2010 to 2015 at Seoul National University Hospital. Positive and negative cases were defined according to whether vertebral fractures developed during follow-up. The cases were divided into training (n=1,416) and test (n=179) sets. A convolutional neural network (CNN)-based prediction algorithm, DeepSurv, was trained with images and baseline clinical information (age, sex, body mass index, glucocorticoid use, and secondary osteoporosis). The concordance index (C-index) was used to compare performance between DeepSurv and the Fracture Risk Assessment Tool (FRAX) and Cox proportional hazard (CoxPH) models.

RESULTS Of the total participants, 1,188 (74.4%) were women, and the mean age was 60.5 years. During a mean follow-up period of 40.7 months, vertebral fractures occurred in 7.5% (120/1,595) of participants. In the test set, when DeepSurv learned with images and clinical features, it showed higher performance than FRAX and CoxPH in terms of C-index values (DeepSurv, 0.612; 95% confidence interval [CI], 0.571 to 0.653; FRAX, 0.547; CoxPH, 0.594; 95% CI, 0.552 to 0.555). Notably, the DeepSurv method without clinical features had a higher C-index (0.614; 95% CI, 0.572 to 0.656) than that of FRAX in women.

CONCLUSION DeepSurv, a CNN-based prediction algorithm using baseline image and clinical information, outperformed the FRAX and CoxPH models in predicting osteoporotic fracture from spine radiographs in a longitudinal cohort.

Keywords

Deep learningOsteoporotic fracturesRisk assessmentX-rays

Figures

**Fig. 1.** Flow chart of patient selection. SNUH, Seoul National University Hospital; AP, anteroposterior.

**Fig. 2.** The architecture of the deep learning-based survival prediction model. HRNet, high-resolution net; ResNet, residual network; BMI, body mass index; FC, fold change.

**Fig. 3.** Performance of the fracture prediction model in the training set using Cox proportional hazard and DeepSurv methods (A) according to clinical models and (B) analytic methods. Model 1 adjusted for age and sex, model 2 additionally adjusted for body mass index, and model 3 additionally adjusted for the use of glucocorticoids and secondary osteoporosis. The C-index values were as follows: Model 1 (Cox proportional hazard [CoxPH], 0.712; 95% confidence interval [CI], 0.652 to 0.773; DeepSurv without images, 0.765; 95% CI, 0.693 to 0.837; DeepSurv with images, 0.794; 95% CI, 0.760 to 0.828); Model 2 (CoxPH, 0.709; 95% CI, 0.648 to 0.771; DeepSurv without images, 0.737; 95% CI, 0.683 to 0.791; DeepSurv with images, 0.782; 95% CI, 0.755 to 0.810); Model 3 (CoxPH, 0.712; 95% CI, 0.654 to 0.770; DeepSurv without images, 0.740; 95% CI, 0.686 to 0.795; DeepSurv with images, 0.764; 95% CI, 0.739 to 0.789). ^aP<0.05 between groups.

**Fig. 4.** Performance of fracture prediction model in the test set using Cox proportional hazard and DeepSurv methods (A) according to clinical models and (B) analytic methods. Model 1 adjusted for age, model 2 additionally adjusted for body mass index, and model 3 additionally adjusted for the use of glucocorticoids and secondary osteoporosis. The C-index values were as follows: Model 1 (Cox proportional hazard [CoxPH], 0.552; 95% confidence interval [CI], 0.550 to 0.554; DeepSurv without images, 0.568; 95% CI, 0.521 to 0.585; DeepSurv with images, 0.544; 95% CI, 0.481 to 0.607); Model 2 (CoxPH, 0.553; 95% CI, 0.552 to 0.555; DeepSurv without images, 0.558; 95% CI, 0.521 to 0.595; DeepSurv with images, 0.610; 95% CI, 0.576 to 0.644); Model 3 (CoxPH, 0.594; 95% CI, 0.584 to 0.604; DeepSurv without images, 0.433; 95% CI, 0.510 to 0.579; DeepSurv with images, 0.612; 95% CI, 0.571 to 0.653). FRAX, Fracture Risk Assessment Tool. ^aP<0.05 between groups.

그림 출처: PubMed Central 공개 논문 (PMC9449110)

Conflict of interest statement

CONFLICTS OF INTEREST . Jae-Won Lee, Byeong Uk Bae, Jin Kyeong Sung, and Kyu Hwan Jung work in the VUNO. Other authors have no conflict of interest relevant to this article.

그 밖의 정보