Diagnostic Accuracy of 3D Ultrasound and Artificial Intelligence for Detection of Pediatric Wrist Injuries

Abstract

Wrist trauma is common in children, typically requiring radiography for diagnosis and treatment planning. However, many children do not have fractures and are unnecessarily exposed to radiation. Ultrasound performed at bedside could detect fractures prior to radiography. Modern tools including three-dimensional ultrasound (3DUS) and artificial intelligence (AI) have not yet been applied to this task. Our purpose was to assess (1) feasibility, reliability, and accuracy of 3DUS for detection of pediatric wrist fractures, and (2) accuracy of automated fracture detection via AI from 3DUS sweeps. Children presenting to an emergency department with unilateral upper extremity injury to the wrist region were scanned on both the affected and unaffected limb. Radiographs of the symptomatic limb were obtained for comparison. Ultrasound scans were read by three individuals to determine reliability. An AI network was trained and compared against the human readers. Thirty participants were enrolled, resulting in scans from fifty-five wrists. Readers had a combined sensitivity of 1.00 and specificity of 0.90 for fractures. AI interpretation was indistinguishable from human interpretation, with all fractures detected in the test set of 36 images (sensitivity = 1.0). The high sensitivity of 3D ultrasound and automated AI ultrasound interpretation suggests that ultrasound could potentially rule out fractures in the emergency department.

Artificial Intelligence for Radiological Paediatric Fracture Assessment: A Systematic Review

“…In two studies, there was a high risk of bias and applicability concerns regarding patient selection [32; 35]. In one of these [35], a 3-dimensional ultrasound sweep of the distal radius was performed by medical students on a 'convenient sample' of children attending the emergency department with wrist injuries. Patients were neither consecutive, nor randomly sampled therefore it was questionable as to how generalisable the study results could be.…”

“…Almost half of all studies had unclear/moderate concerns regarding applicability of patient selection (4/9, 44.4%) [31; 34; 36; 37] and most had concerns regarding applicability of index test (6/9, 66.7%) [31][32][33][34][35][36]. This was predominantly due to studies imposing strict exclusion criteria in their patient selection (e.g.…”

“…exclusion of patients with healing bones, certain types of fractures, and treatment with cast or surgical correction devices) which would limit the application of the algorithm in clinical practice. In four studies the risk of bias for the reference standard was considered unclear/moderate as the radiology readers were unblinded to the clinical history, which may have in uenced their reporting of ndings and subsequent algorithm performance [33][34][35]. Only two studies reported results for external validation of their algorithm using a dataset which was distinct to the training and validation datasets [17; 30].…”

Physics Driven Domain Specific Transporter Framework with Attention Mechanism for Ultrasound Imaging

“…In the above formulation, DGA might not be required due to the availability of labelled data which would impose sufficient control on the architecture during the training process to extract keypoints relevent to the labels. Additionally, DGA might not be required for a large value of k in the unsupervised Transporter, i.e, number of keypoints detected by KeyNet as the main purpose of DGA was to discard features that were auxillary so that existing keypoints focus more on important features, this claim remains unproven currently as we restricted our studies to a small number of keypoints (10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20) to focus more on remaining hyperparameters of the network. A trained Transporter architecture could also be used to formulate a reliable structural similarity index for a frame pair specific to ultrasounds (as inferred by Figure 12) by setting a reconstruction loss threshold in the Transporter network such that the reconstruction loss obtained would be below a threshold if the frame pair is sufficiently similar to each other and vice-versa.…”

“…Deep Learning (DL) models like convolutional neural networks (CNN) and recurrent neural networks (RNN) have been used used for segmentation and classification in ultrasound images including Lung [12], [13] and Wrist [14]. Recently, CNN models that fuse LP filtered images with the original B-mode image have been used for bone segmentation [15].…”