Deep Learning Assistance Closes the Accuracy Gap in Fracture Detection Across Clinician Types

Area of Science:

• Artificial Intelligence in Radiology
• Deep Learning for Medical Imaging
• Musculoskeletal Imaging Diagnostics

Background:

• Missed fractures are a common diagnostic error in musculoskeletal imaging, leading to delayed treatment and morbidity.
• Deep learning (DL) offers a potential solution by training algorithms to detect fractures, mimicking expert clinician judgment.
• Current DL systems for fracture detection face limitations in anatomic scope and require regulatory approval.

Purpose of the Study:

• To evaluate if a Food and Drug Administration-cleared deep learning (DL) system enhances diagnostic accuracy for fracture detection in adult musculoskeletal radiographs across various clinician types.
• To analyze trends in musculoskeletal radiograph interpretation by different clinician types using Medicare claims data.
• To determine if the DL system provides greater benefit to clinicians with less specialized musculoskeletal imaging training.

Main Methods:

• Utilized Medicare Part B data to assess trends in radiograph interpretation by clinician type (2012-2018).
• Conducted a multiple-reader, multiple-case study involving 24 clinicians assessing 175 unique radiographic cases (12 anatomic regions) under aided and unaided conditions.
• Diagnostic accuracy was measured using area under the curve (AUC), sensitivity, and specificity, with fracture miss rate as a secondary outcome.

Main Results:

• Clinicians using the DL system demonstrated higher diagnostic accuracy (aided AUC: 0.94) compared to unaided (unaided AUC: 0.90).
• Sensitivity and specificity improved with DL assistance (Sensitivity: 90% aided vs. 82% unaided; Specificity: 92% aided vs. 89% unaided).
• Clinicians with limited musculoskeletal imaging training significantly reduced their fracture miss rate (9% aided vs. 20% unaided) when using the DL system, nearing the miss rate of radiologists (10%).

Conclusions:

• The deep learning system enhanced fracture diagnosis accuracy for all clinicians, with a particularly significant benefit for those with limited specialized training.
• Implementing DL systems in musculoskeletal imaging can potentially decrease missed fractures, leading to improved patient outcomes and mobility.
• The study supports the utility of AI-powered tools in augmenting clinical decision-making for fracture detection.