sajjadhadi/YOLOv11-Tumor-Detection

Brain Tumor Segmentation Model with YOLO11

Disclaimer

Important Notice: The Brain Tumor Segmentation Model with YOLO11 is intended ONLY for research purposes. This model is NOT designed or suitable for medical decision-making or clinical use. The data used to train this model may be inaccurate, incomplete, or outdated.

Do NOT use this model to diagnose, treat, or manage any health conditions or illnesses. Always consult a qualified healthcare professional for medical advice and diagnosis. The creators and distributors of this model are not responsible for any consequences arising from the use or misuse of this model in medical decision-making.

Use this model with caution and ensure compliance with all relevant regulations and ethical guidelines in your research. The model's predictions and outputs should be interpreted critically and within the context of established medical knowledge.

Overview

This repository presents a robust brain tumor segmentation model built using the YOLO11 architecture from Ultralytics. Designed for semantic segmentation tasks, this model accurately identifies and delineates brain tumors in medical imaging datasets. Leveraging the power of the YOLO11l-seg variant, it achieves high precision and recall, making it a valuable tool for medical image analysis and research applications.

The model was trained and evaluated on a comprehensive dataset of brain tumor images, demonstrating exceptional performance in detecting and segmenting tumor regions. It supports real-time inference and visualization, providing both bounding box and mask predictions for detailed analysis.

Key Features

• Architecture: Based on YOLO11l-seg, a state-of-the-art segmentation model with 203 layers and 27.6M parameters, optimized for accuracy and efficiency.
• Dataset: Trained on a structured dataset with 1502 training images, 429 validation images, and 215 test images, featuring two primary tumor categories.
• Performance: Achieves a mean Average Precision (mAP50) of 0.938 for bounding box detection and 0.815 for mask segmentation on the test set, with an mAP50-95 of 0.666 (boxes) and 0.329 (masks).
• Speed: Offers fast inference with an average of 19.9ms per image on a Tesla P100 GPU, ensuring scalability for large datasets.
• Visualization: Includes built-in functionality to visualize segmentation predictions, enhancing interpretability of results.

Applications

This model is ideal for researchers and practitioners in medical imaging, particularly those focused on neuro-oncology. It can assist in automated tumor detection, treatment planning, and advancing diagnostic tools. The high accuracy and segmentation capabilities make it suitable for both academic research and clinical workflows.

Usage

The model is pre-trained and ready for inference. It requires the Ultralytics library, which can be installed via pip install ultralytics. The dataset is organized in YOLO format, with images and labels split into training, validation, and test sets, accompanied by a dataset.yaml configuration file.

For optimal performance, inference is recommended on a CUDA-enabled GPU, though CPU support is also available. The model outputs both bounding box coordinates and segmentation masks, which can be visualized for qualitative assessment.

Performance Metrics

• Bounding Box Detection:
  • Precision: 0.902
  • Recall: 0.932
  • mAP50: 0.938
  • mAP50-95: 0.666
• Mask Segmentation:
  • Precision: 0.789
  • Recall: 0.799
  • mAP50: 0.815
  • mAP50-95: 0.329

These metrics reflect the model’s reliability in identifying and segmenting brain tumors across diverse test cases, balancing precision and generalization.

Acknowledgments

This model leverages the Ultralytics YOLO framework and was developed using a brain tumor image dataset tailored for semantic segmentation. Contributions from the open-source community and advancements in deep learning for medical imaging have made this work possible.