Določanje centra rumene pege v angiografskih OCT slikah bolnikov s sladkorno boleznijo

Obreza, Gašper (avtor), Milanič, Matija (mentor), Keber, Črt (komentor)

V zaključni nalogi smo na slikah očesne mrežnice, posnetih s tehniko optične koherenčne tomografije v angiografskem načinu (OCTA), določali center rumene pege. Metoda OCTA nam omogoča neinvaziven vpogled v ožilje in-vivo vzorca ter zato predstavlja novo orodje za pomoč pri zdravljenju in razumevanju bolezni, ki vplivajo nanj. Sladkorna bolezen je ena od bolezni, ki prizadanejo žilje očesne mrežnice. Bolezensko spremenjenemu žilju zaradi sladkorne bolezni pravimo diabetična retinopatija (DR). Pri DR je poškodovana mikrovaskularna struktura očesa. V nalogi primerjamo velikosti centra rumene pege med zdravimi osebami ter ljudmi, ki imajo DR. Predstavili bomo algoritem, ki smo ga napisali za segmentacijo fovealne avaskularne cone (FAC). Na segmentirane dele slike smo prilagajali elipse, njihove velikosti smo nato primerjali med slikami zdravih in obolelih mrežnic, posnetih v štirih podskupinah, ki jih delimo glede na velikost zajete slike, 3x3 mm$^2$ in 6x6 mm$^2$, ter glede na globino zajemanja slike, globinski in površinski žilni pletež. Zanima nas, ali lahko potrdimo alternativno hipotezo, ki pravi, da obstajajo razlike v velikostih FAC med zdravimi in obolelimi ljudmi. Ničelna hipoteza pravi, da sprememb ni. Za preverjanje alternativne hipoteze uporabimo statistični test vsote rangov. Vrednost p slik posnetih na površinskem žilnem pletežu pri velikosti zajemanja 3x3 mm$^2$ je $p_{3x3} = 0,0088$ in $p_{6x6} = 0,0129$ za 6x6 mm$^2$, kar je pod izbrano mejo, zato lahko ovržemo ničelno hipotezo. Ne moramo pa je ovreči pri slikah posnetih v globinskem žilnem pletežu, kjer sta rezultata $p_{3x3} = 0,8818$ za 3x3 mm$^2$ in $p_{6x6} = 0,1984$ za 6x6 mm$^2$. Predvidevamo, da je razlog v večji količini žil v površinskem pletežu na katere DR lahko vpliva. V nadaljevanju bomo predlagali nekaj možnih izboljšav, ki bi podale več informacij o krvnožilni strukturi ter spremembah, ki jih povzroči DR. Omenjenih bo nekaj problemov, nastalih pri segmentaciji določenih OCTA slik in pokazali, kako se jih da rešiti. Namen zaključne naloge je preveriti, ali lahko zaznamo spremembe med skupinami. Na podlagi ugotovitev pa razmisliti, kako izboljšati algoritem določanja FAC, da bi bil sposoben določiti, kateri skupini pripada slika. Tako orodje bi bilo uporabno pri oftalmologiji za zgodnje odkrivanje DR.

• Jezik: Slovenski jezik
• Ključne besede: OCTA, DR, analiza slik
• Vrsta gradiva: Zaključna naloga
• Tipologija: 2.11 - Diplomsko delo
• Organizacija: FMF - Fakulteta za matematiko in fiziko
• Leto izida: 2025
• PID: 20.500.12556/RUL-178198
• COBISS.SI-ID: 258963459
• Datum objave v RUL: 21.01.2026

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: Segmentation of macula’s center in OCT angiography images of diabetic patients

Izvleček

In this bachelor’s thesis, we determined the center of the macula on retinal images captured using the Optical Coherence Tomography Angiography (OCTA) technique. The OCTA method allows for a non-invasive view of the vasculature in in-vivo samples, thus representing a new tool to aid in the treatment and understanding of the diseases that affect it. Diabetes is one of the diseases that affect the retinal vasculature. The pathological changes to the vasculature caused by diabetes are known as diabetic retinopathy (DR). In DR, the microvascular structure of the eye is damaged. In this thesis, we compare the size of the macular center between healthy individuals and people with DR. We will present an algorithm we developed for the segmentation of the foveal avascular zone (FAC). We fitted ellipses to the segmented parts of the image and subsequently compared their sizes between images of healthy and diseased retinas. These images were captured in four subgroups, divided by scan size (3x3 mm$^2$ and 6x6 mm$^2$) and by imaging depth (the deep and superficial vascular plexuses). We aim to test the alternative hypothesis, which states that there are differences in FAC sizes between healthy and diseased individuals. The null hypothesis states that no such changes exist. To test the alternative hypothesis, we use a rank-sum statistical test. The p-value of images taken on the superficial vascular plexus with a capture size of 3x3 mm$^2$ is $p_{3x3} = 0.0088$ and $p_{6x6} = 0.0129$ for 6x6 mm$^2$, which is below the chosen limit, so we can reject the null hypothesis. However, we cannot reject it for images taken in the deep vascular plexus, where the results are $p_{3x3} = 0.8818$ for 3x3 mm$^2$ and $p_{6x6} = 0.1984$ for 6x6 mm$^2$. We assume that the reason lies in the larger amount of veins in the superficial plexus that DR can affect. Furthermore, we will propose several possible improvements that could provide more information about the vascular structure and the changes caused by DR. Some problems that arose during the segmentation of certain OCTA images will be mentioned, and we will show how they can be resolved. The purpose of this thesis is to verify whether changes can be detected between the groups. Based on the findings, we will consider how to improve the FAC detection algorithm so that it would be able to determine which group an image belongs to. Such a tool would be useful in ophthalmology for the early detection of DR.

• Ključne besede: OCTA, DR, image analysis