Enhancing Radiographic Diagnosis: CycleGAN-Based Methods for Reducing Cast Shadow Artifacts in Wrist Radiographs.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Cast shadows in radiographs can obscure anatomical details, potentially hindering diagnosis.
• Existing AI techniques, like CycleGAN, have shown promise in addressing this issue.

Purpose of the Study:

• To enhance generative adversarial network (GAN) models for cast shadow reduction in radiographs.
• To evaluate the diagnostic performance and clinical utility of AI-generated, cast-free radiographic images.

Main Methods:

• Retrospective collection of 11,500 adult and pediatric wrist radiographs.
• Enhancement of CycleGAN with perceptual loss and self-attention for cast suppression.
• Qualitative assessment by radiologists on 20 clinical cases and comparison with follow-up imaging reports.

Main Results:

• AI-generated images improved radiologist diagnostic confidence and led to more decisive reports.
• Radiologists could not distinguish AI-enhanced images from unenhanced ones.
• Diagnoses from AI-enhanced images closely matched follow-up imaging reports.

Conclusions:

• AI-driven cast suppression is a clinically meaningful technique for augmenting radiographic interpretation.
• Potential benefits include reduced patient dose, improved efficiency, and fewer repeat imaging procedures.