JMIR Publications

ABSTRACT

Background:

Total hip replacement (THR) is considered the gold standard of treatment for refractory degenerative hip disorders. Identify patients who should receive THR in short term is important for some conservative treatment like intraarticular injection few months before THR may resulted in higher odds of arthroplasty infection, delayed THR after functional deterioration resulted in a poorer outcome and long waiting time to THR for those indicated. Deep learning (DL) in medical image applications which has recently obtained significant breakthroughs. DL in practical wayfinding such as short-termed THR prediction is still lacking.

Objective:

We propose a DL-based assistant system on patients with the pelvic radiograph (PXR) to identify the need for THR within three months.

Methods:

We developed a convolutional neural networks-based DL algorithm to analyze PXRs and to predict the hip region of interest (ROI) whether to have THR or not. The dataset was collected from August 2008 to December 2017. The images included 3013 surgical hip ROIs which underwent THR and 1630 non-surgical hip ROIs without following hip surgery. The images were split by using split-sample validation into training (80%, n=3903), validation (10%, n=476) and testing sets (10%, n=475) to evaluate the algorithm performance.

Results:

The algorithm called SurgHipNet yielded an area under the receiver operating characteristic curve (AUROC) of 0.994 (95% CI, 0.990–0.998). The accuracy, sensitivity, specificity, and F1 score of the model were 0.977, 0.920, 0932, and 0.944, respectively.

Conclusions:

The proposed approach has demonstrated that SurgHipNet showed its ability and possibility to provide efficient support in clinical decision-making and assist physicians in judging the timing to perform THR promptly.