Automatic Detection of Brain Metastases in T1-Weighted Construct-Enhanced MRI Using Deep Learning Model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain metastasis (BM) significantly impacts patient survival and quality of life.
• Accurate BM detection is crucial for effective radiation therapy planning.
• Manual diagnosis of BM is challenging due to small size and varied numbers, necessitating AI assistance.

Purpose of the Study:

• To develop an AI-assisted method for accurate and efficient brain metastasis detection.
• To reduce false positive results in AI-based BM detection while maintaining high accuracy.
• To alleviate the diagnostic workload for radiation oncologists.

Main Methods:

• A modified YOLOv5 deep learning algorithm was developed for BM detection.
• Incorporated a convolutional block attention model to enhance feature map analysis.
• Introduced an additional prediction head for small BM detection and a Swin transformer block to differentiate BMs from cerebral vessels.

Main Results:

• The proposed method achieved a precision of 0.612 and a recall of 0.904.
• Demonstrated superior performance compared to existing methods, with a significant reduction in false positives.
• The F2-score index was utilized to optimize the confidence threshold for detection.

Conclusions:

• The modified YOLOv5 algorithm effectively improves the accuracy and efficiency of brain metastasis detection.
• The AI tool shows potential in reducing the burden on radiation oncologists during clinical diagnosis.
• This advancement aids in better patient management for those with brain metastases.