Skin Lesion Classification to detect Melanoma using Deep Convolution Networks
- 0 Collaborators
Skin Cancer is one of the major health problem with almost 5 million newly diagnosed cases every year throughout the world. Melanoma is the deadliest form of skin cancer and is responsible for almost 10000 deaths each year in just United States. System that can help screen melanoma will greatly help in its diagnosis. ...learn more
Project status: Under Development
Overview / Usage
Definitions:
Melanoma – malignant skin tumor, derived from melanocytes (melanocytic)
Nevus – benign skin tumor, derived from melanocytes (melanocytic)
Seborrheic keratosis – benign skin tumor, derived from keratinocytes (non-melanocytic)
We perform two independent binary classification:
1.Melanoma vs Nevus and Seborrheic keratosis
2. Seborrheic keratosis vs Nevus and Melanoma
This deep learning model can be used by doctors and clinicians to diagnose melanoma with accuracy and minimum cost.
Methodology / Approach
We use dermoscopy images with ground-truth labels as our dataset. The dataset has been provided by International Skin Imaging Collaboration.
As we are using Tensorflow as our deep learning framework, we converted the dataset into TFrecord file for better performance using the Tensorflow input data pipeline. The image data is pre-processed and normalized to be used as input for our learning model. We are using Keras high-level API to develop our network. Currently, we used a primitive Convolution Neural Network to classify the images. Now, we are developing our model and implementing Residual Network (ResNet) to improve the performance. We further plan to augment our data and also add more image pre-processing for better performance.
Technologies Used
Python
Tensorflow
OpenCV
Keras
Numpy
Repository
https://github.com/muditchaudhary/Skin-Lesion-Classification