Scientists from the North Caucasus Federal University (NCFU) have created an innovative neural network for detecting skin cancer, which reduces the likelihood of erroneous diagnoses. This development is an auxiliary diagnostic tool aimed at reducing human influence in medical decision-making and increasing the accuracy of disease detection.

A multimodal neural network system developed at NCFU classifies oncological skin lesions, taking into account unbalanced dermatological data. This innovative approach, as noted by the authors, reduces the number of erroneous predictions by using a modified cross-entropy loss function and analyzing heterogeneous dermatological data with pre-cleaning of hair structures.

According to the results of testing conducted at NCFU, the recognition accuracy of the system in ten diagnostic categories was 85.2%, exceeding the accuracy of visual diagnostics of practicing doctors by 15 percentage points when recognizing pigmented skin lesions, RIA Novosti reports.

It is argued that the Russian system is superior to foreign analogues from Germany, Austria and China. Scientists emphasize that the use of this system by dermatologists as an auxiliary diagnostic method will reduce the human factor, reduce the number of erroneous diagnoses and increase the accuracy of early detection of skin cancer.

The creators plan to further improve the system, including the construction of more complex ensemble systems for neural network analysis of dermatological data.

Previously reportedthat an artificial intelligence model can determine whether you are at high risk of lung cancer.