Browsing by Author "Kaji, Elizabeth S. "

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    A Deep Learning Algorithm to Detect Proximal Humerus Fractures on Radiographs
    (2025) Sperling, John William; Yang, Linjun; Girod, Miguel M. ; Saniei, Sami; Kaji, Elizabeth S. ; Grove, Austin F. ; Khela, Monty; De Marinis Acle, Rodrigo Ignacio; Sanchez Sotelo, Joaquin
    Background Proximal humerus fractures are one of the most common fractures in the elderly. Management of these injuries varies depending on the fracture pattern. It has become recognized that agreement is poor when these fractures are classified according to most traditional schemes. Deep learning (DL) offers the promise to improve recognition of specific fracture patterns. A first, necessary step in the development of DL pipelines to automatically classify fractures into specific patterns is to automatically detect that the proximal humerus is actually fractured, so that a fracture pattern DL classifier is not applied to non-fracture x-rays (normal, arthritic, and other). The purpose of this study was to develop a reliable and trustworthy DL approach to detect proximal humerus fractures on radiographs.MethodsAfter obtaining a patient cohort and reviewing and labeling their associated shoulder images, radiographs of fractured (n=996) and non-fractured (n=607) proximal humerus were used for this study. For model training, a random search was performed to fine-tune and determine the training hyper-parameters. All radiographs were split into six sets. The first five sets were used for model development using five-fold cross-validation, and the sixth set was used for internal model testing. The best-performing model was determined using the F1 score on the sixth set and was further validated using an external test set of 116 separate additional radiographs obtained after proximal humerus fractures. Saliency maps were generated for visual understanding of the DL model.ResultsThe best-performing model from the five-fold cross-validation achieved an accuracy of 0.972 and an F1 score of 0.969 on the hold-out test set. When further validated using the external test set of 116 radiographs of fractured proximal humeri, the model achieved an accuracy/sensitivity of 0.966, misclassifying only 4 of the 116 fractures. The saliency maps showed that the model focused on the perimeter of the humeral head, including the greater tuberosity and the surgical neck, when detecting the fracture.ConclusionThe DL algorithm developed in this study displayed robust and trustworthy performance when detecting the presence of a proximal humerus fracture on radiographs. This algorithm could be implemented in busy emergency department practices so that proximal humerus radiographs with a high probability of fracture can be flagged for dedicated review. Additionally, this algorithm provides the first step to further develop other AI tools to better understand and manage proximal humerus fractures.
  • Thumbnail Image
    Item
    Artificial Intelligence to Automatically Measure on Radiographs the Postoperative Positions of the Glenosphere and Pivot Point After Reverse Total Shoulder Arthroplasty
    (2025) Yang, Linjun; Kaji, Elizabeth S.; Grove, Mr. Austin F.; Marinis Acle, Rodrigo Ignacio de; Velásquez García, Ausberto; Ulrich, Marisa N.; Sperling, Jr. John W.; Marigi, Erick M.; Sánchez-Sotelo, Joaquín
    Introduction: Radiographic evaluation of the implant configuration after reverse total shoulder arthroplasty (rTSA) is a time-consuming task that is frequently subject to interobserver disagreement. Deep learning (DL) artificial intelligence (AI) algorithms have previously demonstrated high accuracy when analyzing relevant angles to determine rTSA distalization and lateralization, as well as glenoid inclination, and humeral alignment. The goal of this study is to build on this existing work to automatically measure the postoperative radiographic location of the glenosphere center of rotation (GCR) and the pivot point in reference to the scapula. Methods: 417 primary rTSA postoperative anteroposterior radiographs were retrieved and utilized for this study. Five measurements were designed and manually performed by three observers: (1) the medial position and (2) the inferior position of the geometric center of rotation of the glenosphere (GCRm and GCRi respectively) relative to the most lateral aspect of the inferior acromion, as well as (3) the projection of the pivot point (PP) to GCR vector on the fossa line (PP projection), (4) the distance between GCR and glenoid (GCR-glenoid distance), and (5) the overall glenoid lateral offset (GLO). Subsequently, a DL algorithm was developed to automatically segment the radiograph and perform the same measurements described above. All measurements were corrected for radiographic magnification using the known glenosphere diameter for each shoulder. Intraclass Correlation Coefficients (ICC) were calculated to assess inter-observer agreements and DL-human agreements on all measurements. Results: The DL algorithm achieved an average Dice Coefficient of 0.86, indicating good segmentation accuracy. The ICCs (95% CI) amongst human observers were 0.86 (0.81-0.90) for the GCRm, 0.93 (0.9-0.95) for the GCRi, 0.95 (0.92-0.96) for the PP projection, 0.85 (0.79-0.89) for GCR-glenoid distance, and 0.92 (0.88-0.95) for GLO. The ICCs between the DL-derived measurements and the average of manual measurements were 0.95 (0.92-0.96) for the GCRm, 0.90 (0.84-0.93) for the GCRi, 0.96 (0.94-0.98) for the PP projection, 0.91 (0.87-0.94) for GCR-glenoid distance, and 0.92 (0.88-0.95) for GLO. The DL algorithm automatically analyzed each testing image in 2 seconds. Conclusions: The developed DL algorithm can automatically measure the location of the glenosphere geometric center of rotation and the location of the pivot point on postoperative radiographs obtained after primary rTSA. Agreement between DL-derived measures and those from human observers was high. This DL algorithm adds to the armamentarium of tools available for automatic assessment of final implant position on radiographs after rTSA.