A two-step neural network-based guiding system for obtaining reliable radiographs for critical shoulder angle measurement

Abstract

Background: The critical shoulder angle (CSA) has been reported to be highly associated with rotator cuff tears (RCTs) and an increased risk of RCT re-tears. However, the measurement of the CSA is greatly affected by the malpositioning of the shoulder. To address this issue, a two-step neural network-based guiding system was developed to obtain reliable CSA radiographs, and its feasibility and accuracy was evaluated.

Methods: A total of 1,754 shoulder anteroposterior (AP) radiographs were retrospectively acquired to train and validate a two-step neural network-based guiding system to obtain reliable CSA radiographs. The study included patients aged 18 years or older who underwent X-rays and/or computed tomography (CT) scans of the shoulder. Patients who had undergone shoulder surgery, had a confirmed fracture, or were diagnosed with a musculoskeletal tumor or glenoid defect were excluded from the study. The system consisted of a two-step neural network that in the first step, localized the region of interest of the shoulder, and in the second step, classified the radiography according to type [i.e., 'forward' when the non-overlapping coracoid process is above the glenoid rim, 'backward' when the non-overlapping coracoid process is below or aligned with the glenoid rim, a ratio of the transverse to longitudinal diameter of the glenoid projection (RTL) ≤0.25, or a RTL >0.25]. The performance of the model was assessed in an offline, prospective manner, focusing on the sensitivity and specificity for the forward, backward, RTL ≤0.25, or RTL >0.25 types (denoted as Sens_{F, B, -, +} and Spec_{F, B, -, +}, respectively), and Cohen's kappa was also reported.

Results: Of 273 cases in the offline prospective test, the Sens_F, Sens_B, Sens_-, and Sens₊ were 88.88% [95% confidence interval (CI): 50.67-99.41%], 94.11% (95% CI: 82.77-98.47%), 96.96% (95% CI: 91.94-99.02%), and 95.06% (95% CI: 87.15-98.40%), respectively. The Spec_F, Spec_B, Spec_-, and Spec₊ were 98.48% (95% CI: 95.90-99.51%), 99.55% (95% CI: 97.12-99.97%), 95.04% (95% CI: 89.65-97.81%), and 97.39% (93.69-99.03%), respectively. A high classification rate (93.41%; 95% CI: 89.14-96.24%) and almost perfect agreement (Cohen's kappa: 0.903, 95% CI: 0.86-0.95) were achieved.

Conclusions: The guiding system can rapidly and accurately classify the types of AP shoulder radiography, thereby guiding the adjustment of patient positioning. This will facilitate the rapid obtainment of reliable CSA radiography to measure the CSA on proper AP radiographs.

Keywords: Computational neural networks; feasibility studies; shoulder radiography.

Figure 1

A flowchart showing how the data sets used for training, internal validation, and the offline prospective test were obtained. The deep-learning models’ results were compared to those of two human radiologists. CT, computed tomography; RTL, ratio of the transverse to longitudinal diameter of the glenoid projection; DRRs, digitally reconstructed radiographs.

Figure 2

Suter’s classification system was used to assess the viewing perspective of shoulder anteroposterior radiographs. (A) Forward: no intersection between the upper glenoid rim (indicated by the red dotted lines) and the coracoid process (indicated by the yellow dotted lines); the coracoid process is positioned either below the upper glenoid rim, or its superior edge is in alignment with the upper glenoid rim. (B) Backward: no overlap is observed between the upper glenoid rim and the coracoid process, indicating that the coracoid process is positioned superior to the upper glenoid rim. (C,D) The upper glenoid rim and the coracoid process overlap, or the coracoid process’s inferior edge is aligned with the upper glenoid rim. RTL = cd versus ab is more than 0.25 (C) or less than 0.25 (D). RTL, ratio of the transverse to longitudinal diameter of the glenoid projection; cd, the transverse diameter; ab, the longitudinal diameter of the glenoid projection.

Figure 3

The framework of the two-step model hierarchical architecture. Both original and DDRs were input for training and validation. The first-step model localized and cropped the glenoid by producing a segmentation map. The red and yellow dotted lines indicate the contours of the glenoid rim and acromion, respectively. After localization by the first-step network, based on Suter’s classification system, the second-step network classified the images into the following types: forward, backward, RTL ≤0.25, or RTL >0.25. DDRs, digitally reconstructed radiographs; RTL, ratio of the transverse to longitudinal diameter of the glenoid projection.

Figure 4

The 4×4 confusion matrix displays the case numbers of the prediction and imaging diagnoses obtained from the offline prospective test using the two-step hierarchical architecture. The right y-axis gradient bar in the confusion matrix represents the number of predicted samples that belong to each class; the color intensity increases as the number of samples increases. RTL, ratio of the transverse to longitudinal diameter of the glenoid projection.