THE MALIGNANT MELANOMA PROJECT


Introduction

The occurrence of malignant melanoma (cancer in birthmarks) is rapidly rising. In Denmark the figure has more than 10-folded during the last 50 years. In the 1940's around 50 occurrences of malignant melanoma were seen yearly, whereas today's figure is around 800.

Malignant melanoma is the most lethal form of skin cancer and usally hits people with the age of 20-50 years. However patients may be cured, if their lesion is detected at an early stage. It is hoped, that computer imaging may assist and improve the detection of such early lesions.

In this project, which was started in collaboration with the Department of Reconstructive Surgery S at the National University Hospital of Denmark in September 1994, we wish to develop methods for detecting malignant melanoma by analyzing color images as well as dermatoscopic images of skin lesions. The approach here is to extract features from an image describing shape, color and other properties of a skin lesion and use an artificial neural network to classify the lesion.


Examples of Skin Lesions

Figure 1: Benign Nevi. Figure 2: Dysplastic Nevi. Figure 3: Malignant Melanoma.

Figure 1 shows a healthy skin lesion. This lesion has no particular risk of developing cancer. In figure 2 a dysplatic nevi is shown. This is a skin lesion, that has a higher risk of developing cancer and is therefore often said to be a precursor to cancer. Figure 3 shows a melanoma. This lesion has cancer.


Preliminary Results

By using color and texture properties extracted from normal color images of skin lesions, 90% of all malignant melanomas in two testsets have been detected.


Links to Melanoma Resources

  • An Introdution to Skin Cancer
  • Dermatologic Image Atlas at Erlangen, Germany
  • Dermatologic Image Database at University of Iowa


    For more info contact: Mads Hintz-Madsen , Lars Kai Hansen


    Back to the Neural Network home page


    © 1996 by Section for DSP, IMM
    No material obtained from this homepage may be used commercially without permission.

    Last modified: March 11, 1996. Comments and suggestions to Mads Hintz-Madsen.