Acute Lymphoblastic Leukemia Detection Using FCM Segmentation + EfficientNetB4 + XGBoost

This project presents a hybrid machine learning pipeline for detecting Acute Lymphoblastic Leukemia (ALL) from blood smear images.
It combines:

• Fuzzy C-Means (FCM) segmentation
• EfficientNetB4 deep feature extraction
• GPU-accelerated XGBoost classification
• Interpretability via qualitative error analysis

###🎯 Final Test Accuracy: 95.53%

---

##📌 1. Dataset

• Blood Cell Cancer (ALL 4-Class)
  https://www.kaggle.com/datasets/mohammadamireshraghi/blood-cell-cancer-all-4class
• Used only for research / academic / educational purposes.

Classes converted into:

• 0 → Benign
• 1 → Malignant (Early Pre-B, Pre-B, Pro-B)

---

##📌 2. Pipeline Overview

Raw Image ↓ FCM Segmentation ↓ EfficientNetB4 Feature Extraction ↓ XGBoost Classifier ↓ Benign / Malignant Prediction

---

📌 3. Features of the Model

✔ Unsupervised segmentation using Fuzzy C-Means
✔ Pretrained EfficientNetB4 for robust features
✔ XGBoost for efficient binary classification
✔ GPU acceleration
✔ Training & validation curves
✔ Confusion matrix + metrics
✔ TP / FP / TN / FN qualitative visualizations

---

📌 4. How to Run

Install Dependencies

pip install -r requirements.txt

Run the Notebook

Open:

segmentation_and_classification.ipynb

All steps are included for segmentation, feature extraction, training, and evaluation.

---

📌 5. Results

✔ Final Accuracy: 95.53%

The model prints a detailed classification report including:

• Precision
• Recall
• F1-score
• Support

A qualitative visualization of predictions (TP, FP, FN, TN examples)
has been saved as:

results/qualitative_analysis.png results/segmentation_examples/

📌 6. File Structure

ALL_Leukemia_Detection_Model/ │ ├── results/ │ ├── qualitative_analysis.png │ └── segmentation_examples/ │ ├── src/ │ ├── fcm_segmentation.py │ ├── feature_extraction.py │ ├── xgboost_classifier.py │ └── utils.py │ ├── segmentation_and_classification.ipynb ├── requirements.txt ├── LICENSE └── README.md

---

📌 7. License

Licensed under MIT License (see LICENSE file).

---

📌 8. Disclaimer

This project is for research and educational purposes only.
It is NOT a clinical diagnostic tool.

---

📌 9. Author

Tagore
Machine Learning Student
Open to internships & freelance ML work.