AI-Assisted Brain Hemorrhage Detection for Early Diagnosis 🧠

Status: Ongoing (Oct 2025 – June 2026)
Institution: Namal University, Mianwali Supervised by: Dr. Ahmed Salim

📝 Abstract

This project focuses on the development of an AI-supported detection system to identify brain hemorrhages from Non-Contrast CT (NCCT) scans and classify their subtypes. Addressing the critical shortage of radiologists in Pakistan's public hospitals, this system aims to reduce diagnosis delays and prevent data loss through a centralized local server solution.

🎯 Project Goals

1. Binary Classification: Detect the presence of hemorrhage (Normal vs. Abnormal).
2. Subtype Classification: Identify the specific type of hemorrhage (e.g., Intracranial, Subdural).
3. Deployment: Provide a user-friendly Web Interface for doctors using Streamlit.
4. Data Management: Implement a local server solution to address the lack of PACS in rural hospitals.

🛠️ Tech Stack

• Language: Python
• Deep Learning: PyTorch / TensorFlow / Keras
• Computer Vision: OpenCV
• Data Processing: Pandas, NumPy, Scikit-learn
• Web Interface: Streamlit, FastAPI
• Hardware: Trained on High-Performance Computing (HPC) GPU resources[cite: 732].

📊 Methodology

The project follows a hybrid deep learning approach:

1. Data Collection: Collaboration with DHQ Hospital Mianwali to collect ~4,800 DICOM images (Hemorrhage & Healthy).
2. Preprocessing: Anonymization, Normalization, and Resizing of DICOM files.
3. Annotation: Expert labelling by radiologists to mark hemorrhage regions and subtypes.
4. Model Architecture: A Hybrid CNN (for feature extraction) combined with lightweight classifiers (ANN/SVM) for classification.
5. Evaluation Metrics: Sensitivity, Specificity, Precision, F1 Score, and AUC.

👥 Team Members

• Uzair Bilal (Lead: Preprocessing, Training, Deployment)
• Muskan Aman Khan (Lead: Data Collection, Annotation, Testing)

🤝 Collaboration

This research is conducted in collaboration with the Department of Diagnostic Imaging, DHQ Hospital Mianwali.