Skin Cancer Detection Project: A Machine Learning Approach Using Deep Learning

Overview

This project focuses on developing an machine learning solution for detecting skin cancer using deep learning techniques. Leveraging pre-trained convolutional neural network (CNN) models, this project aims to provide a accurate and efficient method for dermatological diagnosis.

Highlights

• Models Used: DenseNet121, InceptionV3, Xception
• Accuracy Achieved: DenseNet121 (96%), InceptionV3 (76%), Xception (93%)
• Dataset: HAM10000 (10,000 dermoscopic images)

Comparison of Models Accuracy, F1-Score and Macro Average ROC AUC

Comparison of Average Accuracy of All Classes in percentage

Training and Validation Loss/ DENSENET 121

Training and Validation Accuracy/ DENSENET 121

Features

• Transfer Learning: Implemented transfer learning by fine-tuning the last layers of DenseNet121, InceptionV3, and Xception models, and adding custom classification layers for precise skin cancer classification.
• Data Preprocessing: Comprehensive preprocessing of the HAM10000 dataset, including image resizing and normalization, to ensure optimal training conditions and model performance.
• Model Evaluation: Thorough evaluation using accuracy metrics to compare model performance, highlighting DenseNet121's superior accuracy.

Innovation and Impact

This project demonstrates the potential of deep learning in medical diagnostics, achieving competitive accuracy rates. By leveraging pre-trained models and fine-tuning them for specific tasks, it showcases the adaptability and power of CNNs in solving complex medical challenges.

Real-World Application

• Mobile Integration: Developed a user-friendly mobile application using the Flutter framework, integrating TensorFlow Lite for real-time inference on dermoscopic images captured via camera or uploaded from the gallery. This app enhances accessibility to skin cancer detection, providing instant results and recommendations for further medical evaluation. 🔗

Technical Skills

• Languages: Python, Dart
• Frameworks and Tools: TensorFlow, Keras, Flutter, VS Code, GitHub, Anaconda, Android Studio

Conclusion

This Skin Cancer Detection Project not only highlights the effectiveness of deep learning in medical applications but also demonstrates practical implementation through a mobile app. The high accuracy rates and real-time inference capabilities make it a valuable tool for aiding dermatological diagnosis.

Contributors:

• Aman Prakash Kanth
• Giriraj Garg
• Ashutosh Kumar
• Shubh Raj

Disclaimer:

This application is for educational and informational purposes only. It is not intended to provide medical advice or diagnosis. Always consult a qualified healthcare professional for medical advice and diagnosis.

Feedback and Contributions:

Feedback and contributions are welcome! Feel free to open an issue or submit a pull request with any improvements or suggestions.