20.500.12556/RUL-186220 Globoke nevronske mreže za klasifikacijo spektralno-prostorskih podatkov FTIR Deep neural networks for classification of spectral-spatial FTIR data Ko zdravnik pri diagnozi raka pod mikroskopom pregleduje vzorec tkiva, da bi v celicah prepoznal bolezenske spremembe, mora biti vzorec prej kemično obarvan, kar postopek upočasni in vzorec trajno spremeni. V tej nalogi obravnavamo alternativo barvanju: klasifikacijo tkiva na podlagi infrardeče spektroskopije (angl. Fourier-transform infrared spectroscopy, FTIR), pri kateri vsak piksel namesto barve nosi celoten spekter, iz katerega je mogoče sklepati na kemično strukturo vzorca in iz nje na vrsto tkiva. Taki podatki so veliki, spektralni in prostorski hkrati (spektralna slika), zato je naravna izbira uporaba globokih nevronskih mrež, ki lahko izkoristijo obe vrsti informacije. Replicirali smo enega prvih znanstvenih člankov, ki za ta problem uporabi konvolucijske nevronske mreže; primerja več modelov in pokaže, da prostorska informacija izboljša napovedi, ne objavi pa izvorne kode niti vseh podrobnosti za zanesljivo ponovitev rezultatov. V nalogi predlagane modele čim zvesteje ponovimo na dveh fizično ločenih tkivnih rezinah, sistematično razrešimo manjkajoče podrobnosti in dodamo lastne izboljšave. Primerljivi rezultati z izvornim člankom potrjujejo enakovredne modele, med njimi najkompleksnejšega, prostorsko-spektralnega, ki doseže klasifikacijsko točnost 81,56 % ± 3,18 (deset ponovitev) proti člankovim 79,18 % ± 1,34. When a doctor examines a tissue sample under a microscope to diagnose cancer, the sample must first be chemically stained, which slows the process and permanently alters the tissue. This thesis addresses a staining-free alternative to that: classifying tissue from infrared spectroscopy (FTIR), where each image pixel carries a full spectrum instead of a color, from which the tissue type can be inferred. Such data is large and both spectral and spatial (a spectral image), which makes deep neural networks, able to exploit both kinds of information at once, a natural approach. We replicated one of the first scientific articles to apply convolutional neural networks to this problem; it compares several machine-learning models for the task, and its main contribution is that spatial information improves predictions, but it releases neither source code nor all the details needed to reliably reproduce its results. We recreate the proposed models as faithfully as possible on a proper split between two physically separate tissue samples, systematically resolve the missing details, and complement the solution with our own improvements. Results comparable to those of the original article show that we succeeded in building equivalent models, including the most complex proposed one, the spatial-spectral classification model, which reaches a classification accuracy of 81.56% ± 3.18 (ten repetitions) against the article’s 79.18% ± 1.34. FTIR spektroskopija globoke nevronske mreze klasifikacija tkiva konvolucijske nevronske mreze metoda podpornih vektorjev FTIR spectroscopy deep neural networks tissue classification convolutional neural networks support vector machines true false false Slovenski jezik Angleški jezik Diplomsko delo/naloga 2026-08-28 12:10:34 2026-08-28 12:10:47 2026-08-31 10:19:34 0000-00-00 00:00:00 2026 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 38915 289431043 Jesenk_Svit_-_Globoke_nevronske_mreze_za_klasifikacijo_spektralno-prostorskih_podatkov_FTIR.pdf Jesenk_Svit_-_Globoke_nevronske_mreze_za_klasifikacijo_spektralno-prostorskih_podatkov_FTIR.pdf 1 EC3C016717C9ED16AC5B707BC391FD61 87b55af8fc08b2ff62d2786923b7ed81580962c38bd62db62767cbaf5fc6d01e 52f5a3bc-a2c8-11f1-9b0d-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=245336 Fakulteta za računalništvo in informatiko 0 0 0