Uporabniška učinkovitost odprtokodnih platform pri segmentaciji računalniško tomografskih slik : diplomsko delo

Kovač Šumah, Sara (author), Đuričić, Aleksandar (author), Žibert, Janez (mentor), Kaučič, Aleš (comentor), Alukić, Erna (reviewer)

Uvod: Segmentacija predstavlja ključni del analize slikovnih podatkov in omogoča natančno prepoznavanje anatomskih ali patoloških struktur na CT slikah. Na tem področju obstajajo različni pristopi, ki vključujejo ročne in avtomatske metode, kjer se uporabljajo postopki umetne inteligence, med njimi globoko učenje. Njihova temeljna komponenta so nevronske mreže. Za potrebe segmentacije je na trgu na voljo več različnih programskih rešitev. Namen: Želeli smo sistematično predstaviti različne metode segmentacije v medicinski slikovni diagnostiki in primerjati uporabniško učinkovitost komericalnega programskega orodja Syngo.via s prostodostopnima orodjema 3D Slicer in MONAI Label. Metode dela: V teoretičnem delu diplomskega dela smo uporabili deskriptivno metodo s pregledom domače in tuje literature na področju umetne inteligence in medicine. Drugi del naloge pa je bil eksperimentalen, ki smo ga praktično izvedli v različnih programskih orodjih, v katerih smo opravili segmentacijo anatomskih struktur na CT posnetku prsnega koša in primerjali praktično uporabnost med njimi ter uporabniško izkušnjo. Rezultati: Ročna segmentacija zahteva precej spretnosti in znanja, hkrati pa je tudi časovno potratna. Syngo.via je, kljub svoji uporabniku izjemno prijazni naravi, omejena s funkcijami v primerjavi z drugimi naprednimi programskimi orodji za segmentacijo CT slik. Natančnost segmentacije v MONAI Labelu je ustrezna in primerljiva z ročno segmentacijo, hkrati pa zahteva manj uporabnikove interakcije, manj znanja ter se izvaja neprimerljivo hitreje kot ročna segmentacija. Segmentacija v 3D Slicer-ju je bila zahtevnejša kot ob uporabi MONAI Label, saj je zahtevala več uporabnikove interakcije in poznavanja anatomije, kar prav tako vpliva na časovno trajanje postopka, ki je v vseh primerih bolj zamudno kot pri MONAI Labelu. Razprava in zaključek: Syngo.via predstavlja visoko napreden program za obdelavo in analizo medicinskih slik, ki izstopa po svoji izjemni zmogljivosti in nezahtevni uporabi. Povezana je z visokimi stroški in omejenostjo glede novih funkcionalnosti, ki jih je potrebno nadgrajevati in doplačevati, saj je v lasti podjetja Siemens Healthineers. Nasprotno pa odprtokodne platforme, kot sta 3D Slicer in MONAI Label, nudijo brezplačno uporabo, širšo dostopnost in sodelovanje v raziskovalni skupnosti ter transparentnost delovanja. S tem omogočajo ustvarjanje natančnih modelov, ki se lahko prilagodijo specifičnim potrebam in izzivom različnih kliničnih ali raziskovalnih primerov.

• Language: Slovenian
• Keywords: diplomska dela, radiološka tehnologija, segmentacija, računalniška tomografija, Syngo.via, MONAI, 3D Slicer
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: ZF - Faculty of Health Sciences
• Place of publishing: Ljubljana
• Publisher: [S. Kovač Šumah : A. Đuričić]
• Year: 2023
• Number of pages: 38 str.
• PID: 20.500.12556/RUL-159595
• UDC: 616-07
• COBISS.SI-ID: 201751555
• Publication date in RUL: 13.07.2024

Secondary language

• Language: English
• Title: User efficiency of open-source platforms in the segmentation of computed tomography images : diploma work

Abstract

Introduction: Segmentation represents a crucial part of the analysis of medical image data and enables precise identification of anatomical or pathological structures in CT images. In this field, various approaches exist, including manual and automatic methods, where artificial intelligence procedures, among deep learning, are used.Their fundamental component is neural networks. For the purpose of segmentation, there are several different software solutions available on the market. Purpose: We aimed to systematically present various segmentation methods in medical imaging diagnostics and compare the user effectiveness of the commercial software tool Syngo.via with open-source tools 3D Slicer and MONAI Label. Methods: In the theoretical part, we employed a descriptive method by reviewing domestic and foreign literature in the field of artificial intelligence and medicine. The second part was experimental and was practically conducted using various software tools. In these tools, we performed segmentation of anatomical structures on a chest CT scan and compared their practical usability and user experience. Results: Manual segmentation requires considerable skill and knowledge and is also time-consuming. Despite its extremely user-friendly nature, Syngo.via is limited in functionalities compared to other advanced software tools for segmenting CT images. The segmentation accuracy in MONAI Label is adequate and comparable to manual segmentation while requiring less user interaction, less knowledge, and is executed incomparably faster than manual segmentation. Segmentation in 3D Slicer was more challenging than using MONAI Label, as it required more user interaction and knowledge of anatomy, which also affects the duration of the process, which is more time-consuming in all cases compared to MONAI Label. Discussion and conclusion: Syngo.via represents a highly advanced program for processing and analyzing medical images, standing out for its exceptional performance and user-friendly nature. However, it is associated with high costs and limitations regarding new functionalities that need to be upgraded and paid for, as it is owned by Siemens Healthineers. On the contrary, open-source platforms like 3D Slicer and MONAI Label offer free usage, broader accessibility, and participation in the research community, as well as transparency of operation. This enables the creation of precise models that can be adapted to the specific needs and challenges of various clinical or research cases.

• Keywords: diploma theses, radiologic technology, segmentation, computer tomography, Syngo.via, MONAI, 3D Slicer