Computer-aided detection of brain metastasis on 3D MR imaging: Observer performance study

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Brain metastasis (BM) detection on MRI is crucial for patient management.
• Accurate interpretation of brain MR imaging requires high diagnostic performance from radiologists.
• Computer-aided detection (CAD) systems offer potential to enhance diagnostic accuracy.

Purpose of the Study:

• To evaluate the impact of a computer-aided detection (CAD) system on radiologists' ability to detect brain metastasis (BM) using 3D brain MR imaging.
• To assess diagnostic performance metrics including sensitivity, false positives, and reading time.
• To determine if CAD improves diagnostic accuracy compared to unassisted interpretation.

Main Methods:

• Development of a CAD system for BM detection using template-matching, K-means clustering, and artificial neural networks.
• Retrospective analysis of MR images from 140 patients (110 with BM, 30 without BM).
• Four reviewers interpreted images with and without CAD, with results analyzed using JAFROC methodology.

Main Results:

• The CAD system achieved 87.3% sensitivity with 302.4 false positives per case.
• Overall diagnostic performance significantly improved with CAD (FOM 0.874 vs. 0.898, p < 0.01).
• Less-experienced reviewers showed statistically significant improvement (FOM 0.834 vs. 0.877, p < 0.01), with an average additional review time of 72 seconds.

Conclusions:

• Computer-aided detection (CAD) serves as an effective second reader for radiologists interpreting brain MR imaging.
• CAD enhances diagnostic performance in detecting brain metastasis (BM), particularly benefiting less-experienced reviewers.
• The integration of CAD systems can improve the accuracy and efficiency of brain metastasis diagnosis.