Avelios successfully develops new algorithm for classifying skin diseases

Originally published by the Technical University of Munich (14.01.2022). The summary below links back to the full article at the source.

Researchers at the Technical University of Munich have developed FusionM4Net, a new artificial intelligence system designed to classify skin diseases with greater accuracy than previous algorithms. The innovation centers on an improved data fusion technique that integrates multiple information sources used by dermatologists.

The multi-modal approach

The algorithm processes three distinct data types simultaneously:

• Clinical photographs of skin lesions
• Microscopic images of suspicious skin areas
• Patient metadata (age, gender, and similar attributes)

Distinguishing features

The system's effectiveness comes from its multi-stage design. As PD Dr. Tobias Lasser from the Munich Institute of Biomedical Engineering explains, the algorithm combines image data first, then integrates patient metadata in sequential steps rather than processing everything simultaneously like competing systems.

Performance results

FusionM4Net achieved an average diagnostic accuracy of 78.5 percent, surpassing all comparable algorithms tested. The research team trained the system using publicly accessible datasets to ensure reproducibility.

Clinical implementation

The research team collaborates with the Dermatology and Allergology Clinic at Ludwig Maximilian University of Munich to adapt the algorithm for real-world clinical use, accounting for variations in data availability across different medical facilities. The code is freely available at ciip.in.tum.de/software.html.

Source: tum.de