Deep-learning and radiomics ensemble classifier for false positive reduction in brain metastases segmentation

Area of Science:

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

Background:

• Stereotactic radiosurgery (SRS) is standard for brain metastases (BMs).
• Manual segmentation of multiple BMs (mBMs) is time-consuming.
• Automated segmentation platforms can generate false positives.

Purpose of the Study:

• To develop and validate a deep-learning and radiomics ensemble classifier to reduce false positives in automated BMs segmentation.
• To improve the efficiency and accuracy of treatment planning for mBMs patients undergoing SRS.

Main Methods:

• A Siamese network and a radiomic-based support vector machine (SVM) classifier were combined into an ensemble model.
• The Siamese network identifies inter-class differences, while the SVM classifies segmentations using radiomic features.
• The ensemble model was integrated into an existing BMs segmentation platform.

Main Results:

• The ensemble classifier achieved high performance with an accuracy of 0.91, sensitivity of 0.96, specificity of 0.90, and AUC of 0.93.
• Integration into the platform significantly improved the false positive over the union (FPoU) from 0.55 to 0.09.
• The false negative rate (FNR) was maintained at 0.07, while the average segmentation FNR was 0.13.

Conclusions:

• The proposed deep-learning and radiomics ensemble classifier effectively reduces false-positive segmentations in BMs.
• This integrated tool enhances the accuracy and efficiency of SRS treatment planning for patients with mBMs.
• The developed method offers a beneficial advancement for managing mBMs in clinical SRS workflows.