325 KB23 KB202220252025Joveski, Lukaštevilka:4str. 67-70COBISSID_HOST:248534275ISSN:2820-5014URN:URN:NBN:SI:doc-1V0O29TQslFakulteta za farmacijo, ŠSSFDPlaceboconvolutional neural networks (CNN)cutaneous melanomadeep learningdermatoskopijadermoscopydiagnosisDiagnostikagloboko učenjekonvolucijske nevronske mreže (CNN)kožni melanomMelanomDiagnostika melanoma z novimi "deep-learning" metodami| vpliv in diagnostika dermatoveneroloških bolezni|Melanoma is one of the most dangerous forms of skin cancer, with early detection significantly improving survival rates. Traditional diagnostic methods, such as clinical examination and dermoscopy, are reliable but subjective and dependent on the experience of the physician. Advances in artificial intelligence, particularly deep learning, offer new possibilities for automated, accurate, and faster melanoma diagnosis. Deep neural network models, trained on extensive image datasets, have already achieved comparable or even better accuracy than dermatologists. However, several challenges remain, such as imbalanced data, lack of metadata, ethical dilemmas, and difficulty transferring models to clinical practice. The integration of genetic and clinical data of the patients, along with the development of lightweight and robust architectures, can contribute to more personalised and widely accessible patient care through advanced deep learning methods. The aim of this review is to present general deep learning approaches in dermoscopy of cutaneous melanoma, to evaluate their effectiveness, and to highlight the key advantages, challenges, and opportunities for clinical applicationKožni melanom je ena najnevarnejših oblik kožnega raka, katerega zgodnje odkritje bistveno poveča možnosti za preživetje. Tradicionalne diagnostične metode, kot sta klinični pregled in dermatoskopija, so zanesljive, a subjektivne in odvisne od izkušenosti zdravnika. Napredek na področju umetne inteligence, zlasti globokega učenja, ponuja nove možnosti za avtomatizirano, natančno in hitrejšo diagnostiko kožnega melanoma. Modeli globokih nevronskih mrež, usposobljeni na obsežnih slikovnih zbirkah, so že dosegli primerljivo ali celo boljšo natančnost kot dermatologi. Kljub obetavnim rezultatom pa ostajajo številni izzivi, kot so neuravnoteženi podatki, pomanjkanje metapodatkov, etične dileme in težave pri prenosu modelov v klinično prakso. Integracija genetskih in kliničnih podatkov pacientov ter razvoj lahkih in robustnih arhitektur lahko pripomoreta k bolj personalizirani in široko dostopni obravnavi pacientov s pomočjo naprednih »deep-learning« metod. Namen tega pregleda je predstaviti splošne pristope globokega učenja v dermatoskopiji kožnega melanoma, oceniti njihovo učinkovitost ter izpostaviti ključne prednosti, izzive in priložnosti za klinično uporaboTEXTznanstveno časopisjejournalsSlovenian National E-content AggregatorNational and University Library of Slovenia