Application of modern computer methods for the analysis of intercellular interaction for improving the diagnosis and surgical treatment of laryngeal squamous cell carcinoma

[S. Valjarević]

88 str.

Medicinske nauke - Rekonstruktivna hirurgija / Medical Sciences - Reconstructive Surgery

Histopatološka dijagnostika promena na sluznici larinksa može biti zahtevna kada je u pitanju razlikovanje displastičnih lezija od početnog stadijuma karcinoma, procena prisustva malignih ćelija u graničnim preparatima uzetim sa tkiva koje makroskopski izgleda zdravo, kao i analiza tkiva prilikom sumnje na recidiv bolesti. Teksturalna analiza predstavlja savremenu računarsku metodu koja može da evaluira početne promene u međućelijskoj interakciji, obliku ćelije i strukturi hromatina, a koje nisu uočljive konvencionalnim mikroskopskim metodama. Cilj ovog istraživanja bio je da se utvrdi senzitivnost računarskih algoritama teksturalne analize u detekciji strukturnih promena karakterističnih za ćelije i tkivo planocelularnog karcinoma larinksa. Na preparatima sluznice larinksa kod 100 pacijenata (50 sa malignitetom, 50 sa hroničnim laringitisom) napravljeno je 2000 reprezentativnih digitalnih mikrografa tkiva korišćenjem optičkih instrumenata, a potom su mikrografi analizirani primenom metoda baziranih na primeni teksturalne analize. Rezultati našeg istraživanja ukazuju na to da postoji značajna razlika između tkivne citoarhitekture planocelularnog karcinoma larinksa i hroničnog laringitisa u pogledu teksturalnih karakteristika kvantifikovanih GLCM tehnikom, kao i u pogledu indikatora vejvlet matematičke analize. Morfološki slična jedra epitelnih ćelija u tkivu planocelularnog karcinoma larinksa i tkivu kod hroničnog laringitisa pokazuju značajne razlike u vrednostima teksturalnih i vejvlet indikatora. Moguće je razviti modele mašinskog učenja zasnovane na logističkoj regresiji, slučajnim šumama i višeslojnom perceptronu za razlikovanje malignog od benignog tkiva koristeći teksturalne i vejvlet indikatore kao ulazne podatke. Teksturalna i vejvlet matematička analiza ćelija i tkiva, kao i modeli mašinskog učenja bazirani na njima, imaju potencijal da u budućnosti, nakon dodatne evaluacije i razvoja, budu uključeni u standardne protokole u otorinolaringologiji za dijagnozu i hirurško lečenje planocelularnog karcinoma larinksa.

Histopathological diagnosis of changes in the laryngeal mucosa can be challenging when it comes to distinguishing dysplastic lesions from the initial stage of carcinoma, assessing the presence of malignant cells in borderline specimens taken from tissue that appears macroscopically healthy, and analyzing tissue in cases of suspected disease recurrence. Textural analysis represents a contemporary computer-based method that can evaluate initial changes in intercellular interaction, cell shape, and chromatin structure that may not be apparent with conventional microscopic methods. The aim of this research was to determine the sensitivity of computer algorithms in textural analysis for detecting structural changes characteristic of cells and tissue in squamous cell carcinoma of the larynx. Using optical instruments, 2000 representative digital micrographs of laryngeal mucosa specimens from 100 patients (50 with malignancy, 50 with chronic laryngitis) were obtained and then analyzed using methods based on the application of textural analysis. The results of our research indicate a significant difference in the tissue cytoarchitecture between squamous cell carcinoma of the larynx and chronic laryngitis in terms of textural characteristics quantified by the GLCM technique, as well as in terms of wavelet mathematical analysis indicators. Morphologically similar nuclei of epithelial cells in the tissue of squamous cell carcinoma of the larynx and tissue in chronic laryngitis show significant differences in the values of textural and wavelet indicators. It is possible to develop machine learning models based on logistic regression, random forests, and multilayer perceptron to differentiate between malignant and benign tissue using textural and wavelet indicators as input data. Textural and wavelet mathematical analysis of cells and tissues, as well as machine learning models based on them, have the potential, after additional evaluation and development, to be incorporated into standard protocols in otolaryngology for the diagnosis and surgical treatment of squamous cell carcinoma of the larynx in the future.

planocelularni karcinom larinksa, teksturalna analiza, međućelijska interakcija, računarske metode, modeli mašinskog učenja

laryngeal squamous cell carcinoma, textural analysis, intercellular interaction, computer methods, machine learning models

616.22-006.6-089:004(043.3)

Tekst