FALS-YOLO: An Efficient and Lightweight Method for Automatic Brain Tumor Detection and Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Brain tumors pose significant threats to neurological health and survival.
• Magnetic Resonance Imaging (MRI) is crucial for brain tumor diagnosis and treatment.
• Existing YOLO-based models struggle with multi-scale feature extraction and computational demands in brain tumor MRI analysis.

Purpose of the Study:

• To develop an improved, lightweight model for brain tumor detection and instance segmentation in MRI images.
• To address the limitations of existing models in multi-scale feature extraction and computational efficiency.

Main Methods:

• Proposed FALS-YOLO model based on YOLOv8n-Seg, incorporating FLRDown, AdaSimAM, and LSCSHN modules.
• FLRDown enhances multi-scale perception; AdaSimAM suppresses noise and improves feature fusion; LSCSHN achieves high-precision segmentation with reduced parameters.
• Experiments conducted on the tumor-otak dataset.

Main Results:

• FALS-YOLO achieved high performance in detection (Precision: 0.892, Recall: 0.858, mAP@0.5: 0.912) and segmentation (Precision: 0.899, Recall: 0.863, mAP@0.5: 0.917).
• Outperformed YOLOv5n-Seg, YOLOv8n-Seg, YOLOv9s-Seg, YOLOv10n-Seg, and YOLOv11n-Seg.
• Compared to YOLOv8n-Seg, FALS-YOLO reduced parameters by 31.95%, computational load by 20.00%, and model size by 32.31%.

Conclusions:

• FALS-YOLO offers an efficient, accurate, and practical solution for automatic brain tumor detection and instance segmentation.
• The model is particularly suitable for environments with limited computational resources.
• The proposed modules effectively enhance multi-scale feature perception and segmentation precision while reducing computational burden.