Dermatological Disease Diagnosis using color-skin images


This project is a machine intervention in medical diagnostics. Etymologically, dermatology is the medical discipline of analysis and treatment of skin anomalies. The system presented is a machine intervention in contrast to human arbitration into the conventional medical personnel based ideology of dermatological diagnosis. The system works on two dependent steps - the first detects skin anomalies and the latter identifies the diseases. The system operates on visual input i.e. high resolution color images and patient history. In terms of machine intervention, the system uses color image processing techniques, k-means clustering and color gradient techniques to identify the diseased skin. For disease classification, the system resorts to feedforward backpropagation artificial neural networks. The system exhibits a diseased skin detection accuracy of 95.99% and disease identification accuracy of 94.016% while tested for a total of 2055 diseased areas in 704 skin images for 6 diseases.

System Workflow

Sytem Description

The system is skin tone independent that is the robustness of the system is such that it can detect and classify skin diseases on divergent skin tones e. g. dark, fair, pale etc. The system is built upon color images of diseased skin (as opposed to the popular grayscale paradigm) as input and minimal external features that require human intervention since the idea is to automate the process to the maximum ex tent possible. The methodology is divided into several sequential steps and they are image acquisition, color gradient generation to find anomalies on skin, clustering and labeling of the regions of interest, feature extraction, system training and system testing.

Brief Methodology

Firstly, images were acquired using a high resolution digital camera from a reputed hospital - Sir Salimullah Medical College and Mitford Hospital, Dhaka. A reference object was needed in order to maintain coherence of the zooming of images and actual size of regions of interest. Secondly, in order to detect the anomalies on regular skin – the images were filtered through several spatial filtering methods to reduce noise (i.e. hair, blur etc.) and then the color gradients were generated which can be looked at as a novel edge detection mechanism that is applied to RGB color images rather than its predecessor – the binary or grayscale images. Consequently the anomalies are detected from the color gradients of the images. Thirdly, upon detecting the regions of interest (anomalies), binary masks were generated by clustering the points/pixels that belong to the diseased parts of skin and labeled them for further classification. The fourth step is feature extraction. The color means and standard deviations were calculated for all 3 channels (RGB) of the images for both the diseased and regular skin of the patients from the dataset, a feed-forward back-propagation neural network was trained and the complete dataset was tested using tenfold cross validation for precise and legitimate accuracy and performance analysis for classification.

Comments are closed.