Back to all projects
Skin Cancer Detection
AI/ML

Skin Cancer Detection

Project Overview

A deep learning image classification system using TensorFlow and CNN to categorize skin lesions into 9 different classes with ~90% accuracy.

Project Details

Release DateNov 2024
TechnologyPythonTensorFlowCNN
Live Preview GitHub Code

Skin Cancer Classifier: Deep Learning Lesion Analysis

This project is a medical computer vision application designed to analyze skin lesions and classify them into 9 distinct diagnostic categories, supporting early clinical screening.

Deep Learning Pipeline

  • **Neural Architecture:** Implemented a Convolutional Neural Network (CNN) in TensorFlow/Keras with 5 convolutional layers, Max Pooling, and Dropout regularization.
  • **Data Augmentation:** Configured data pipelines to resize inputs to 180x180 pixels and apply random rotations, zooms, and horizontal flips, combating model overfitting.
  • **Inference Client:** Built a streamlined Streamlit application where users upload photographs of skin markings and receive percentage confidence bars.

    • Solving Model Performance Limitations

  • **Imbalanced Dataset Optimization:** Applied focal loss functions and class weight balancing during training, boosting rare class recall from 64% to 88% while maintaining overall diagnostic accuracy at ~90%.