Malaga university researchers develop tool based on artificial intelligence to improve detection of melanomas

A research team at the University of Malaga (UMA) is using artificial intelligence to help detect skin cancer. In its paper published in Computers in Biology and Medicine, the experts used an approach based on genetic algorithms and automatic image classification to analyse and detect melanoma in a set of photos.

The aim of the scientists is to gain a better understanding of how the artificial intelligence (AI) algorithm 'thinks' to solve a problem or perform a specific task.

However, they pointed out, "not all algorithms are transparent, meaning that the people who are using them cannot know why they make the decisions they do, nor what steps the system has followed to obtain a result".

To explain how the algorithm 'thinks', the experts developed an explainer. This highlights on the photos the areas on which the system focuses to detect melanoma. Visually, it shows how different pixels light up in yellow, helping experts identify which area the algorithm is looking at and what process it follows to determine whether a spot is a melanoma or not.

To test it, the Khaos research group collected a set of medical images showing spots, moles, freckles and other skin marks. Typically, doctors look at changes in shape, colour, size or texture to determine whether they are benign or have progressed to melanoma.

The researchers found that when the algorithm got it right, it looked at the same characteristics, while when it failed, it looked at less representative areas. "It detects different skin lesions such as melanoma, nevus, keratosis and healthy moles," said José Manuel García-Nieto, a researcher at the University of Málaga.

Early diagnosis

The experts were then assisted by Dr Miguel Ángel Berciano from the Intercentre Oncology Unit of the Virgen de la Victoria University Hospital (Malaga), who, checked the algorithm's responses and guided the researchers on the veracity of its answers, and on the areas of interest for detecting possible melanoma in skin marks.

The researchers found that the explainer works and added that, based on it, "the algorithm can be taught to follow the appropriate steps to detect the most complicated melanoma cases".

"The programme in no way replaces the work and diagnosis of a doctor, but we propose it as a tool for pre-diagnosis of melanoma, as is already done in other hospitals with the artificial intelligence that detects breast cancer," García-Nieto said.