An international research team has developed a deep-learning algorithm that they say classifies skin lesions more accurately than alternative programs. This new algorithm, developed by an international team and the Munich Institute of Biomedical Engineering at the Technical University of Munich in collaboration with the Department of Dermatology and Allergology at the University Hospital of LMU Munich, incorporates data from multiple sources, including clinical images, dermoscopic images of the skin lesion, and patient metadata. It is still under development.
Called FusionM4Net, the algorithm can also provide multi-label skin classification, and differentiate five different categories of skin lesions. The developers say their multi-stage approach distinguishes FusionM4Net from other algorithms, which typically merge all available data in one step. FusionM4Net first uses the available image data combined with the patient’s metadata, then weights that information in the decision-making process.
According to a study published in Medical Image Analysis (Feb. 2022. DOI: 10.1016/j.media.2021.102307), the authors say their approach increases the average accuracy of the algorithm to 77.0% and the diagnostic accuracy to 78.5%.
“Future routine clinical use of algorithms with high diagnostic accuracy might help ensure that rare diseases are also detected by less experienced physicians and it might mitigate decisions affected by stress or fatigue,” said lead researcher Dr. Tobias Lasser in the press release.
The developers note that the code for FusionM4Net is freely available at (https://ciip.in.tum.de/software.html).