Impact of Data Presentation on Physician Performance Utilizing Artificial Intelligence-Based Computer-Aided Diagnosis...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Ultrasound (US) is crucial for detecting breast cancer, but differentiating benign from malignant lesions remains challenging.
• This limitation leads to suboptimal positive predictive value for biopsy (PPV3) and reduced specificity.
• Artificial intelligence (AI) decision support (DS) systems show potential to enhance US diagnostic performance.

Purpose of the Study:

• To evaluate workflow integration schemas for AI-based DS in breast US.
• To assess the impact of different reading methodologies (sequential vs. independent) on diagnostic accuracy.
• To measure the effect of DS integration on inter-operator variability.

Main Methods:

• The study compared diagnostic accuracy using two distinct workflow schemas for AI-DS integration.
• Two reading methodologies, sequential and independent, were assessed.
• Statistical analysis included area under the receiver operating curve (AUC) and Kendall's tau-b for inter-operator agreement.

Main Results:

• Significant differences in diagnostic accuracy were observed between the evaluated workflow schemas.
• The integration method notably affected inter-operator variability, as indicated by Kendall's tau-b.
• The findings highlight the importance of workflow design in AI implementation.

Conclusions:

• Workflow schema significantly impacts the diagnostic performance of AI-DS in breast US.
• Optimizing workflow integration is critical for maximizing the benefits of AI in clinical practice.
• This evaluation provides practical insights for deploying AI technologies in diagnostic environments.