The use of new technologies, artificial intelligence and deep learning is becoming more and more common in all segments of human life. Artificial intelligence and its capabilities are increasingly finding their place in medical diagnostics. Methods and models for working with and processing medical images are finding a special place in medical diagnostics. In this study, the performance of four deep learning models was analyzed for the super-resolution task on dermatological images. The models were tasked with scaling images from 32 \(\times \) 32 to 128 \(\times \) 128, demonstrating strong performance in image reconstruction and confirming their effectiveness in the field. For x4 scaling, the models confirmed their performances. The best results achieved SRCNN and ESPCN models with value of PSNR above 30 and SSIM above 0.82. The EDSR and SRGAN achieved results for the PSNR above 28 and above 0.70 for SSIM, which confirmed state-of-the-art and their complexity compared to other two CNN-based methods. On the other hand, this paper also shows and explains how metrics such as PSNR and SSIM affect the assessment of model quality, but also highlights their limitations in the context of human perception, which is especially important when evaluating the results of models such as SRGAN.