MONDAY, Jan. 31, 2022 (HealthDay News) -- A new multimodal learning algorithm can perform multilabel classification of skin lesions, according to a study published in the February issue of Medical Image Analysis.
Peng Tang, from the Technical University of Munich, and colleagues proposed a novel two-stage multimodal learning algorithm for multilabel classification of skin diseases. A FusionNet, which exploits and integrates the representation of clinical and dermoscopy images at the feature level, was constructed at the first stage; a Fusion Scheme 1 was used to conduct the information fusion at the decision level. To further incorporate patients' meta-data, a Fusion Scheme 2 was proposed, which integrates the multilabel predictive information from the first stage and patients' meta-data information. The final diagnosis was formed by fusion of the predictions from both stages. The algorithm was assessed on the 7-point checklist dataset for multilabel classification of skin disease.
The researchers found that the proposed FusionM4Net first stage achieved an average accuracy of 75.7 percent for multiclassification tasks and 74.9 percent for diagnostic tasks, without using the patient's meta-data; this was more accurate than other methods. The second stage of FusionM4Net, which included patients' meta-data increased average accuracy to 77.0 percent and diagnostic accuracy to 78.5 percent.
"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," a coauthor said in a statement.