Abstract

To date, skin cancer is the most frequently diagnosed form of oncopathology. Visual diagnosis of this cancer type in the early stages is difficult due to similar visual manifestations in benign and malignant types of pigmented formations, and its accuracy depends on the experience of dermatologist. Artificial intelligence technologies are able to equal and surpass accuracy of visual diagnostic methods, but there is a risk of false positive prognosis, when malignant pigmented formation may be recognized as benign. One of the possible ways to improve recognition accuracy and reduce false positive conclusions is simultaneous use of different preprocessing methods, heterogeneous data analysis, and use of modified learning loss functions to eliminate negative impact of unbalanced dermatological data. The study presents an intelligent system for unbalanced heterogeneous dermatological data analyzing based on a multimodal neural network trained using modified cross-entropy loss function. The accuracy of classification of pigmented skin lesions in 10 diagnostically significant categories in the proposed system based on the AlexNet convolutional neural network architecture reached 83.87%. Due to the emerging synergy when using various methods to improve performance quality of intelligent systems, the number of false positive forecasts of the intelligent system was reduced, and the accuracy was increased by 1.31-4.91 percentage points, depending on the convolutional neural network architecture. Implementation of the developed system as a tool for auxiliary diagnostics can reduce the consumption of financial and labor resources involved in the medical industry, as well as increase the chance of early detection of pigmentary oncopathologies.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call