New tech taps deep learning to improve skin cancer diagnoses

The combination of human experience and AI can contribute to a drastic improvement in diagnostic accuracy in early skin cancer detection, stakeholders say, with the potential for almost 100 per cent accuracy.
Jeff Rowe

Every year, GPs in Australia alone see more than a million patients per year for skin cancer, but providers Down Under and around the world are getting some help in battling skin cancer with a new mole analyzer that uses new AI technology to help tell the difference between harmless moles and dangerous melanomas.

“The earlier skin cancer is detected, the better the prognosis. The leisure behavior of sunbathing in many parts of the world makes early detection of skin cancer more important worldwide,”  explained Global Brand Director Kathrin Niemela of FotoFinder Systems, which recently unveiled the new technology. 

Specifically, Moleanalyzer pro is a portal that lets physicians confirm their skin cancer diagnosis using evaluation techniques, combining specialist expertise with AI and including the option of receiving a second opinion from international skin cancer experts.

Using deep learning, FotoFinder Systems first calculates and compares size, diameter and structure of moles and quantifies their percentage deviations.  Its Convolutional Neural Network was ‘trained’ with a large data collection of dermoscopic images and corresponding diagnoses. 

Through growing experience and its own autonomous rules, it then distinguishes between benign and malignant lesions.  The analysis then determines a risk assessment score of both melanocytic and non-melanocytic skin lesions, allowing physicians to verify their diagnoses.  

“Moleanalyzer pro features the possibility to manually evaluate lesions according to acknowledged checklists and optionally contains an innovative algorithm based on AI, allowing a risk-of-malignancy evaluation,” Niemela said. “In the last few years, the new algorithm has been trained with a large number of dermoscopic images. FotoFinder Systems has an international network of partners who contribute to the training of the algorithm with their pictures of histologically proven lesions.” 

According to Niemela, a man-against-machine study involving 58 dermatologists from 17 nations found that whereas the experts correctly identified 86.6 of malignant skin tumors, Moleanalyzer pro successfully detected 95 per cent.  The technology also identified 82.5 per cent of benign naevi correctly, while the experts identified 71.3 per cent as benign.

“The AI represents a ‘silent virtual colleague’ that delivers a virtual opinion simply, uncomplicatedly and at any time,” Niemela said. “But together with the human experience delivered by the optional second opinion service, the tool helps to increase diagnostic accuracy.” 

FotoFinder Systems is working towards making this AI score available for doctors on mobile devices.