Binary Tumor Classifier with Deep Learning

This repository contains a binary tumor classifier for the Kaggle ISIC 2024 competition, which distinguishes between benign and malignant tumors using images. The project is designed to serve as both a learning resource and a functional model, incorporating detailed explanations of the methodology and technical approaches.

---

Overview

The primary goal of this project is to:

• Develop a baseline model achieving at least 70% accuracy.
• Refine the model to achieve over 95% accuracy through iterative improvements.

The dataset used is a subset of the ISIC Archive, containing fewer than 400 labeled images (malignant or benign), along with a metadata CSV file.

---

Key Features

In this project, you will learn how to:

• Prepare image classification metadata from a CSV file.
• Use data augmentation techniques to address data scarcity.
• Build a CustomDataset class and manage DataLoaders for training and validation.
• Train a neural network using PyTorch to classify tumors with high accuracy.

---

Dataset

The ISIC dataset is available at: ISIC Archive.

Preprocessing Steps:

1. Metadata Preparation:

   • The CSV file includes labels for each image (malignant or benign).

2. Data Augmentation:

   • Techniques such as flipping, rotating, and cropping are applied to expand the dataset and improve generalization.

---

Model Architecture

The baseline model is built using PyTorch and follows this architecture:

1. Convolutional Neural Network (CNN) backbone for feature extraction.
2. Fully connected layers for binary classification.

Further architectural enhancements include:

• Use of pretrained models (e.g., ResNet or EfficientNet).
• Optimization through hyperparameter tuning and advanced loss functions.

---

Training and Evaluation

Training:

• Data is split into Training and Validation sets.
• The model is trained using:
  • Cross-Entropy Loss as the loss function.
  • Adam Optimizer for adaptive learning rates.

Evaluation:

Metrics:

• Accuracy: Ensures baseline goal (≥ 70%) is achieved.
• Precision, Recall, and F1-Score: Assess the model's robustness.
• A confusion matrix is used to evaluate predictions.