Skip to Content

Leukemia Detector

Leukemia Detector provides clinical-grade diagnostic support by using advanced Deep Learning to classify leukemia subtypes with 97.7% accuracy, featuring automated image validation and secure patient record management. 

Leukemia Detector: AI-Powered Hematology Analysis

Leukemia Detection is a cutting-edge medical imaging platform that uses Deep Learning to identify and classify Acute Lymphoblastic Leukemia (ALL). By analyzing blood smear images with high-precision neural networks, the system assists medical professionals in distinguishing between healthy cells and three specific malignant subtypes, ensuring faster and more accurate diagnostic support.

The Challenge: Diagnostic Bottlenecks

  • Manual Labor: Traditional diagnosis requires expert hematologists to manually examine blood slides, which is time-consuming and exhausting.

  • Subjectivity: Human fatigue can lead to inconsistent results or the misclassification of leukemia subtypes.

  • Limited Access: Many regions lack the specialized pathologists needed for rapid early detection, which is critical for patient survival.

The Solution: Automated Precision

The system provides a reliable, automated second opinion. By processing blood smear images through an advanced "electronic brain," it can instantly validate if a cell is a white blood cell and then categorize it with nearly 98% accuracy.

Key Capabilities

  • Subtype Identification: Goes beyond a simple "yes/no" by identifying specific leukemia stages (Early Pre-B, Pre-B, and Pro-B).

  • Smart Input Validation: Automatically rejects non-medical images or low-quality uploads using histogram-based "template matching."

  • High-Speed Analysis: Provides real-time predictions via a secure web interface.

  • Clinical Record Keeping: Integrated database to track patient history and diagnostic results over time.

  • Benchmarked Accuracy: Utilizes the most effective AI architectures (including EfficientNet and ResNet) to ensure clinical-grade reliability.

How It Works

  1. Capture: A digital image of a peripheral blood smear is uploaded to the system.

  2. Verify: The AI first confirms the image is actually a white blood cell to prevent errors.

  3. Analyze: The Deep Learning model examines the cell's features at a microscopic level.

  4. Report: The system delivers a classification result along with a confidence score and saves the record for the physician.

Results & Impact

  • Exceptional Accuracy: The custom-engineered model achieved a 97.7% success rate in correctly identifying leukemia stages.

  • Consistency: Unlike manual review, the AI provides the same rigorous analysis every time, regardless of workload.

  • Robustness: Trained on thousands of clinical images from 89 different patients to ensure it works across diverse cases.

My Role as Lead AI Engineer

I developed the entire end-to-end pipeline, from the initial data science research to the final clinical deployment.

  • Model Engineering: Designed and benchmarked multiple AI architectures to find the highest-performing model for medical use.

  • System Architecture: Built the Flask API and MySQL integration to turn a complex AI model into a functional, user-friendly tool.

  • Quality Control: Engineered the image validation system to ensure that only relevant medical data is processed by the neural networks.

  • Data Optimization: Implemented advanced "augmentation" techniques to train the AI to be more accurate across different lighting and slide conditions.