🧠 Brain Tumor Detection with YOLOv8 + SAM 🔬

An advanced object detection and segmentation pipeline that leverages YOLOv8 for tumor localization and SAM (Segment Anything Model) for precise mask generation. Developed using PyTorch, OpenCV, and integrated with an optional Streamlit UI for real-time interaction.

---

📌 Overview

This project combines the power of YOLOv8 (by Ultralytics) and SAM (by Meta AI) to detect and segment brain tumors from MRI scans.

• 🔍 Detects multiple objects with YOLOv8
• 🧠 Generates pixel-accurate tumor masks using SAM
• 💾 Outputs bounding box and segmentation coordinates
• 📊 Supports general object detection from COCO too

---

🧪 Sample Outputs

🎯 Input MRI Scan

Original brain scan used as input:

✅ Output: Detected Tumor(s)

After YOLOv8 detection + SAM segmentation:

---

🚀 How It Works

1. YOLOv8 detects all bounding boxes of interest (e.g., tumors).
2. Each bounding box is passed to SAM, which generates a detailed segmentation mask.
3. Both bounding boxes and polygonal masks are saved in .txt files for further use or annotation.

---

🧰 Installation

Clone the repository and install the necessary libraries:

git clone https://github.com/your-username/brain-tumor-yolo-sam.git
cd brain-tumor-yolo-sam

🔧 Install Python Dependencies:

pip install torch torchvision torchaudio
pip install opencv-python
pip install ultralytics
pip install git+https://github.com/facebookresearch/segment-anything.git

▶️ (Optional) Install Streamlit:

pip install streamlit

---

🏃 Run the Project

➤ Run YOLO + SAM Pipeline:

cd YOLOV8_SAM
python detect_multi_object_SAM.py

➤ (Optional) Run Streamlit App:

streamlit run appy.py

---

📦 Models Used

Model	Description	Source
YOLOv8s	Object detection for bounding boxes	Ultralytics YOLOv8
SAM ViT-H	Segmentation from bounding boxes	Meta AI Segment Anything

---

📚 Dataset

• BraTS 2021: Brain Tumor Segmentation Challenge
• MRI-based brain tumor scans
• Converted to YOLOv8 format inside yolo_brain_dataset/
• Masks saved in output/masks/

---

🔢 Output Files

File Name	Description
bounding_box_image1.txt	YOLOv8 bounding boxes for test image
yolo_mask_image1.txt	Normalized mask polygon coordinates
output_detection.jpg	Annotated image with box + mask overlay

---

📷 Tested on COCO Too!

The pipeline also works on general object detection tasks. For example:

• Detected: Class 74 = Book

You can try feeding other images to test its generalizability.

---

👨‍💻 Developed By

Abdul Rafay
📚 BS Software Engineering | 🎯 AI & ML Enthusiast
🔗 LinkedIn

---

📜 License

This repository is licensed under the MIT License.

---

🌟 Support & Contribution

If you found this helpful:

• ⭐ Star the repo
• 🍴 Fork it and contribute
• 📢 Share on LinkedIn and tag me!

---

🔍 Accurate detection. 🎯 Precise segmentation. 🚀 Built with passion.