Skin Cancer Detection

Project Domain / Category

Image Processing/AI/Web App

Abstract / Introduction

Skin cancer is one of the top causes of death globally, and melanoma is one of its most dangerous forms. It accounts for nearly 75% of skin cancer deaths and represents about 4% of all deaths worldwide. Early detection of melanoma can lead to successful treatment, otherwise it has the potential to spread to other parts of the body, making expert diagnosis essential. To diagnose melanoma, doctors first analyse dermoscopic images of the affected skin and then confirm their findings with an expert. However, even highly experienced doctors can struggle to differentiate between melanoma and normal skin parts because of their visual similarities. Unfortunately, the challenge of early detection remains due to the overlapping visual features of melanoma and non-melanoma conditions. The traditional methods of melanoma detection are highly time-consuming and error prone. This situation demands an automated computer aided diagnosis tools and techniques to detect and diagnose cancer at its early stages.

The project includes detecting skin cancer from dermoscopic images and performed binary classification into melanoma or non-melanoma class. You are required to develop a web app in which the user will enter images and check the status.

Functional Requirements:

•         Dataset Collection: Collect image dataset from available free repositories or any other online source.

•         Pre-Processing: Use different image processing techniques to create a uniform, normalized image dataset. You may need to perform data augmentation in this step.

•         Model Selection: Analyze different deep learning-based CNN models and select a suitable one for classification.

•         Dataset Splitting: Split the dataset into training and testing set for model evaluation.

•         Model Training: Train the selected model using training dataset.

•         Validation and Hyperparameter Tuning: Validate the model's performance using the validation set and fine-tune hyperparameters like learning rate, batch size, and network architecture to achieve the best results.

•         Model Evaluation: Check the performance of the model used using testing dataset and different evaluation metrics.

•         Real-time Detection: Implement a real-time skin cancer detection pipeline using OpenCV to upload an image from and apply the trained model for skin cancer detection.

•         User-Interface: Develop a user-friendly interface in which the user can upload dermoscopic images for analysis. The interface should provide visual feedback, such as original images alongside classification results.