Related Experiment Video
Updated: Jun 7, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.7K
Spectrum of errors in nodule detection and characterization using machine learning: A pictorial essay
Jabi E Shriki1, Ted Selker2, Kristina Crothers3
1University of Washington School of Medicine, United States; Department of Radiology, Puget Veterans Administration Healthcare System, United States.
Current Problems in Diagnostic Radiology
|December 19, 2024
Summary
Computer-aided nodule detection (CAD) software improves accuracy but can cause errors. Radiologists must understand these potential pitfalls in nodule detection, localization, and characterization for safe clinical use.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Computer-aided nodule detection (CAD) software enhances diagnostic performance in academic settings.
- Widespread clinical adoption necessitates understanding CAD system limitations and potential errors.
Purpose of the Study:
- To review the spectrum of errors associated with computer-aided nodule detection (CAD) systems.
- To familiarize radiologists with potential pitfalls encountered in routine clinical practice.
Main Methods:
- Review of clinical practice cases involving CAD software.
- Categorization of errors into detection, localization, and characterization issues.
- Illustrative case examples to demonstrate specific error types.
Main Results:
- Identified errors in nodule detection, including false positives and negatives.
- Observed inaccuracies in nodule localization, affecting precise measurement.
- Encountered challenges in nodule characterization, potentially impacting diagnostic classification.
Conclusions:
- Despite advancements, CAD software can generate errors in detection, localization, and characterization.
- Radiologists must maintain critical review of images, being aware of AI-generated errors.
- Mindful oversight is crucial for safe and effective integration of CAD systems in clinical workflows.
