Automatic detection and multi-component segmentation of brain metastases in longitudinal MRI

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Radiotherapy planning and assessment

Background:

• Manual segmentation of brain metastases (BMs) in MRI is labor-intensive and prone to errors, impacting radiotherapy planning and follow-up.
• Automated methods are crucial for improving the efficiency and accuracy of lesion detection and segmentation.

Purpose of the Study:

• To develop and evaluate an automated approach for detecting and segmenting brain metastases (BMs) in longitudinal MRI scans.
• To enhance segmentation performance by re-segmenting lesions using information from previous time points.
• To perform multi-component segmentation distinguishing enhancing tissue, edema, and necrosis.

Main Methods:

• Utilized a retrospective dataset of 184 contrast-enhanced T1-weighted MRIs from 49 patients with BMs.
• Developed a re-segmentation technique propagating lesion masks from prior scans as an additional input.
• Employed one-tailed t-tests to compare segmentation and detection metrics, with significance set at p < 0.05.

Main Results:

• Achieved a Dice Similarity Coefficient (DSC) of 0.79 and an F1-score of 0.80 for new lesion segmentation.
• Re-segmentation model significantly outperformed the standard segmentation model on follow-up scans (DSC 0.78 vs 0.56, F1-score 0.88 vs 0.60).
• Re-segmentation improved performance for enhancing lesions (DSC 0.76 vs 0.53) and edema (0.52 vs 0.47), while necrosis segmentation remained comparable (0.62 vs 0.63).

Conclusions:

• Automated segmentation of new brain metastases and subsequent re-segmentation in follow-up MRI scans is feasible and effective.
• The re-segmentation approach significantly enhances accuracy, particularly for longitudinal tracking and radiotherapy planning.
• This method overcomes challenges in segmenting small lesions, showing consistent performance regardless of lesion size.