Development and validation of a deep-learning model for detecting brain metastases on 3D post-contrast MRI: a...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Accurate detection of brain metastasis (BM) is crucial for patient management.
• Manual identification of BM is time-consuming and labor-intensive.
• Development of automated systems is needed to improve diagnostic efficiency.

Purpose of the Study:

• To develop, validate, and evaluate a novel brain metastasis detection (BMD) system.
• To assess the performance of the BMD system in terms of detection sensitivity and false positives.
• To compare the diagnostic performance and reading time of radiologists with and without the assistance of the BMD system.

Main Methods:

• A multi-scale cascaded convolutional network was developed using 3D-enhanced T1-weighted MR images.
• The BMD system was trained on a retrospective dataset of 573 patients with BMs and 377 patients without BMs.
• Prospective validation was performed using internal and three external datasets, including analysis of lesion-based detection sensitivity and false positives per patient.

Main Results:

• The BMD system demonstrated high detection sensitivity (95.8% in test set, 96.0% in internal validation, 88.9%-95.5% in external sets).
• BMD system achieved higher detection sensitivity (93.2%) compared to radiologists without BMD (68.5%-80.4%).
• Radiologist detection sensitivity improved with BMD (92.7%-95.0%), and reading time was reduced by 47% for trainees and 32% for experienced radiologists.

Conclusions:

• The developed BMD system enables accurate brain metastasis detection.
• Assisted reading with BMD significantly improves radiologists' detection sensitivity.
• The BMD system effectively reduces radiologist reading times, enhancing diagnostic workflow efficiency.