Implementation of an Artificial Intelligence-Based Double Read System in Capturing Pulmonary Nodule Discrepancy in CT...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Radiology errors, particularly perceptual errors, significantly impact patient outcomes.
• Early detection of pulmonary nodules is crucial for managing lung cancer and metastasis.
• Small pulmonary nodules are frequently missed during routine radiological interpretation.

Purpose of the Study:

• To implement and evaluate an automated double read system for detecting discrepancies in pulmonary nodule reporting.
• To improve patient safety by identifying potentially significant but undocumented pulmonary nodules.

Main Methods:

• Prospective application of machine vision and natural language processing algorithms to CT scans and radiology reports.
• Identification of discrepancies where pulmonary nodules were detected by algorithms but not reported.
• Flagging discrepancies for secondary radiologist review and analysis.

Main Results:

• Processed 4,900 CT studies, identifying 450 potential discrepancies.
• Final review of 104 cases revealed 50 instances of undocumented pulmonary nodules.
• Seven cases with significant undocumented nodules led to report addendums and clinician notification.

Conclusions:

• An automated double read system effectively detects pulmonary nodule reporting discrepancies.
• This system can be implemented securely on-premises, improving patient safety with minimal workflow disruption.
• The technology demonstrates viability for enhancing radiology practices and patient care.