MRI-based two-stage deep learning model for automatic detection and segmentation of brain metastases

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Brain metastases (BMs) are a common complication of cancer, significantly impacting patient prognosis.
• Accurate detection and segmentation of BMs in MRI are crucial for treatment planning and monitoring.

Purpose of the Study:

• To develop and validate a two-stage deep learning model for automated detection and segmentation of brain metastases (BMs) in MRI images.
• To evaluate the model's performance against existing methods, particularly for small BMs.

Main Methods:

• A retrospective study utilized T1-weighted and T1-weighted contrast-enhanced MRI from 649 patients.
• A two-stage deep learning model was developed, comprising a segmentation network for proposal generation and a classification network for false-positive reduction.
• Performance was assessed using sensitivity, precision, F1-score, Dice coefficient, and relative volume difference (RVD).

Main Results:

• The two-stage model achieved 90% sensitivity and 56% precision on the test set.
• It significantly outperformed one-stage methods in detecting small BMs (<5 mm), with 66% sensitivity.
• Segmentation performance included an average Dice score of 81% and RVD of 20%.

Conclusions:

• The developed two-stage deep learning model effectively detects and segments brain metastases.
• The model demonstrates high sensitivity and precision, with notable improvement in identifying small BMs.
• This AI approach offers a promising tool for improving the accuracy and efficiency of BM analysis in clinical practice.