Aims. The purpose of this study was to develop a convolutional neural network (CNN) for fracture detection, classification, and identification of greater tuberosity displacement ≥ 1 cm, neck-shaft angle (NSA) ≤ 100°, shaft translation, and articular fracture involvement, on plain radiographs. Methods. The CNN was trained and tested on radiographs sourced from 11 hospitals in Australia and externally validated on radiographs from the Netherlands. Each radiograph was paired with corresponding CT scans to serve as the reference standard based on dual independent evaluation by trained researchers and attending orthopaedic surgeons. Presence of a fracture, classification (non- to minimally displaced; two-part, multipart, and glenohumeral dislocation), and four characteristics were determined on 2D and 3D CT scans and subsequently allocated to each series of radiographs. Fracture characteristics included greater tuberosity displacement ≥ 1 cm, NSA ≤ 100°, shaft translation (0% to < 75%, 75% to 95%, > 95%), and the extent of articular involvement (0% to < 15%, 15% to 35%, or > 35%). Results. For detection and classification, the algorithm was trained on 1,709 radiographs (n = 803), tested on 567 radiographs (n = 244), and subsequently externally validated on 535 radiographs (n = 227). For characterization, healthy shoulders and glenohumeral dislocation were excluded. The overall accuracy for fracture detection was 94% (area under the receiver operating characteristic curve (AUC) = 0.98) and for classification 78% (AUC 0.68 to 0.93). Accuracy to detect greater tuberosity fracture displacement ≥ 1 cm was 35.0% (AUC 0.57). The CNN did not recognize NSAs ≤ 100° (AUC 0.42), nor fractures with ≥ 75% shaft translation (AUC 0.51 to 0.53), or with ≥ 15% articular involvement (AUC 0.48 to 0.49). For all objectives, the model’s performance on the external dataset showed similar accuracy levels. Conclusion. CNNs proficiently rule out proximal humerus fractures on plain radiographs. Despite rigorous training methodology based on CT imaging with multi-rater consensus to serve as the reference standard, artificial intelligence-driven classification is insufficient for clinical implementation. The CNN exhibited poor diagnostic ability to detect greater tuberosity displacement ≥ 1 cm and failed to identify NSAs ≤ 100°, shaft translations, or articular fractures. Cite this article: Bone Joint J 2024;106-B(11):1348–1360

Artificial intelligence (AI) is, in essence, the concept of ‘computer thinking’, encompassing methods that train computers to perform and learn from executing certain tasks, called machine learning, and methods to build intricate computer models that both learn and adapt, called complex neural networks. Computer vision is a function of AI by which machine learning and complex neural networks can be applied to enable computers to capture, analyze, and interpret information from clinical images and visual inputs. This annotation summarizes key considerations and future perspectives concerning computer vision, questioning the need for this technology (the ‘why’), the current applications (the ‘what’), and the approach to unlocking its full potential (the ‘how’).

Cite this article: Bone Joint J 2022;104-B(8):911–914.

Aims

The aim of this study was to develop and evaluate a deep learning-based model for classification of hip fractures to enhance diagnostic accuracy.

Our aim was to determine if the detection rate of infection of total hip replacements could be improved by examining the removed prostheses. Immediate transfer of prostheses to an anaerobic atmosphere, followed by mild ultrasonication to dislodge adherent bacteria, resulted in the culture of quantifiable numbers of bacteria, from 26 of the 120 implants examined. The same bacterial species were cultured by routine microbiological techniques from only five corresponding tissue samples. Tissue removed from 18 of the culture-positive implants was suitable for quantitative tissue pathology and inflammatory cells were present in all samples. Furthermore, inflammatory cells were present in 87% of tissue samples taken from patients whose implants were culture-negative. This suggests that these implants may have been infected by bacteria which were not isolated by the techniques of culture used. The increased detection of bacteria from prostheses by culture has improved postoperative antibiotic therapy and should reduce the need for further revision

We studied the detection of joint replacements at airport security checks in relation to their weight, using two types of detector arch. A single-source, unilateral detector showed different sensitivities for implants on different sides of a test subject. All implants weighing more than 145 g were detected by one of the arches. The degree of detection was directly related to the logarithm of the weight of the prosthesis in patients, with a linear correlation (r. 2. = 0.61). A bilateral arch detected all prostheses weighing over 195 g. With their usual sensitivity settings many joint replacements were detectable; an identification pass containing the site and weight of such prostheses would help to avoid the need for body-search procedures

Our aim was to determine the clinical value of MRI and CT arthrography in predicting the presence of loose bodies in the elbow. A series of 26 patients with mechanical symptoms in the elbow had plain radiography, MRI and CT arthrography, followed by routine arthroscopy of the elbow. The location and number of loose bodies determined by MRI and CT arthrography were recorded. Pre-operative plain radiography, MRI and CT arthrography were compared with arthroscopy. Both MRI and CT arthrography had excellent sensitivity (92% to 100%) but low to moderate specificity (15% to 77%) in identifying posteriorly-based loose bodies. Neither MRI nor CT arthrography was consistently sensitive (46% to 91%) or specific (13% to 73%) in predicting the presence or absence of loose bodies anteriorly. The overall sensitivity for the detection of loose bodies in either compartment was 88% to 100% and the specificity 20% to 70%. Pre-operative radiography had a similar sensitivity and specificity of 84% and 71%, respectively. Our results suggest that neither CT arthrography nor MRI is reliable or accurate enough to be any more effective than plain radiography alone in patients presenting with mechanical symptoms in the elbow

Visualisation of periacetabular osteolysis by standard anteroposterior (AP) radiographs underestimates the extent of bone loss around a metal-backed acetabular component. We have assessed the effectiveness of standard radiological views in depicting periacetabular osteolysis, and recommend additional projections which make these lesions more visible. This was accomplished using a computerised simulation of radiological views and a radiological analysis of simulated defects placed at regular intervals around the perimeter of a cadaver acetabulum. The AP view alone showed only 38% of the defects over all of the surface of the cup and failed to depict a 3 mm lesion over 83% of the cup. When combined with the AP view, additional 45° obturator-oblique and iliac-oblique projections increased the depiction, showing 81% of the defects. The addition of the 60° obturator-oblique view further improved the visualisation of posterior defects, increasing the rate of detection to 94%. Based on this analysis, we recommend using at least three radiographic views when assessing the presence and extent of acetabular osteolysis

Aims

The aim of this study was to evaluate the ability of a machine-learning algorithm to diagnose prosthetic loosening from preoperative radiographs and to investigate the inputs that might improve its performance.

Aims

The aim of this study was to establish the incidence of developmental dysplasia of the hip (DDH) diagnosed after one-year of age in England, stratified by age, gender, year, and region of diagnosis.

We have examined the accuracy of 143 consecutive ultrasound scans of patients who subsequently underwent shoulder arthroscopy for rotator-cuff disease. All the scans and subsequent surgery were performed by an orthopaedic surgeon using a portable ultrasound scanner in a one-stop clinic. There were 78 full thickness tears which we confirmed by surgery or MRI. Three moderate-size tears were assessed as partial-thickness at ultrasound scan (false negative) giving a sensitivity of 96.2%. One partially torn and two intact cuffs were over-diagnosed as small full-thickness tears by ultrasound scan (false positive) giving a specificity of 95.4%. This gave a positive predictive value of 96.2% and a negative predictive value of 95.4%. Estimation of tear size was more accurate for large and massive tears at 96.5% than for moderate (88.8%) and small tears (91.6%). These results are equivalent to those obtained by several studies undertaken by experienced radiologists.

We conclude that ultrasound imaging of the shoulder performed by a sufficiently-trained orthopaedic surgeon is a reliable time-saving practice to identify rotator-cuff integrity.

We used MRI to examine the hips of 32 asymptomatic patients at 9 to 21 months after renal transplantation covered by high-dose corticosteroids. Five hips in three patients showed changes which indicate avascular necrosis, although radiographs, CT scans and isotope scans were normal. These patients had repeat MRI scans after another two years and three years. One patient with bilateral MRI changes developed symptoms and abnormal radiographs and CT and isotope scans in one hip nine months after the abnormal MRI. Intraosseous pressure was found to be raised in both hips, and core biopsies revealed necrotic bone on both sides. The other three hips have remained asymptomatic with unchanged MRI appearances three years after the initial MRI. It seems that idiopathic avascular necrosis does not always progress to bone collapse in the medium term.

We have assessed the effect of a variety of implants commonly used in fracture fixation and joint replacement on the activation of metal detectors at airport security gates. A volunteer with metal implants strapped on and patients with implants in situ walked through the device. Implants used in fixation do not activate it, except for Richards cannulated screws. An Austin-Moore prosthesis does set off the detector, but a single joint replacement does not. Three or four joint replacements activate the alarm and patients with these implants should be warned of this possibility.