Cervical Cancer Classification Using Deep Learning

This documentation presents the final project output for the Cervical Cancer Classification system. The project uses Deep Learning for feature extraction and a Classical Machine Learning approach for classification.

---

Project Objective

The goal is to classify cervical cell images into two categories:

• Normal — Healthy cervical cells.
• Abnormal — Cells showing precancerous or cancerous traits.

The system uses a pre-trained ResNet152 CNN for feature extraction and a Logistic Regression model for classification.

---

Model Workflow Summary

1. Dataset Preparation

   • Consolidated original 5-class dataset into binary form.
   • Split data into training (80%) and testing (20%) subsets.

2. Feature Extraction

   • Used ResNet152 (without the final FC layer) to generate 2048-D feature vectors for each image.

3. Classifier Training

   • Trained a Logistic Regression model using the extracted features.
   • Applied StandardScaler for normalization.

4. Web Deployment

   • Integrated the model into a Flask web app for user-friendly image predictions.

---

Output & Results

1. Model Performance Metrics

Classification Report:

Metric	Value
Accuracy	98.15%
Precision	98.00%
Recall	98.00%
F1-Score	98.00%

---

2. Confusion Matrix

3. Sample Prediction Output

When a user uploads an image through the web interface:

Sample Prediction

4. File Structure

├─ .gitignore
├─ README.md
├─ requirements.txt
└─ src
   ├─ classical_models
   │  └─ train_classifier.py
   ├─ data_dat_deletion.py
   ├─ data_preperation
   │  ├─ binary_folders.py
   │  ├─ preprocess.py
   │  └─ split_dataset.py
   ├─ feature_extraction
   │  └─ feature_extraction.py
   ├─ predictor
   │  ├─ image_predictor.py
   │  └─ init.py
   ├─ test
   │  └─ test_image.py
   └─ webapp
      ├─ app.py
      └─ templates
         └─ predictor.html

---

Conclusion

This project successfully demonstrates how combining Deep Learning for feature extraction with Classical ML classification yields efficient and interpretable results in medical image analysis. The web-based interface makes the model accessible for non-technical users, contributing toward AI-assisted cervical cancer screening.

---