kth.sePublications KTH
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
AI-baserad segmentering av pulmonella blodkärl inom medicinsk bildanalys
KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems.
KTH, School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH), Biomedical Engineering and Health Systems.
2025 (Swedish)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesisAlternative title
AI-Based Segmentation of Pulmonary Blood Vessels in Medical Image Analysis (English)
Abstract [sv]

Manuell analys av datortomografibilder (CT) utan kontrast, med fokus på exspiratoriska pulmonella blodkärl, är idag en utmaning. För att underlätta och förbättra diagnostiken vid lungsjukdomar som interstitiella lungsjukdomar (ILD), inklusive lungfibros, finns ett växande behov av kvantitativa och automatiserade bildanalysmetoder. Syftet med denna studie var att utveckla och utvärdera en AI-modell för segmentering av lungkärl baserad på nnUNet, ett självkonfigurerande neuralt nätverk för medicinsk bildanalys. Totalt 33 thorax CT-bilder i exspirationsfas utan kontrastmedel hämtades från Karolinska Universitetssjukhuset i Solna via databasen PACS. Bilderna segmenterades i 3D Slicer och användes för träning av modellen med nnUNet i flera stegvisa träningsrundor. Modellens prestanda utvärderades kvantitativt med Dice Similarity Coefficient (DSC) mot manuellt segmenterade referensbilder från sju slumpmässigt utvalda fall. Resultaten visade att modellen successivt förbättrades och kunde identifiera en större del av kärlträdet speciellt i de centrala lungregionerna. Den slutliga modellen efter tre träningsrundor uppnådde ett medelvärde för DSC på 0,9911. Segmenteringen var dock mindre exakt i de perifera kärlområdena. nnUNet visar lovande resultat för automatiserad segmentering av pulmonella kärl i exspiratoriska CT-bilder utan kontrast och har potential för framtida kliniska tillämpningar. Fortsatt utveckling och utökning av träningsdata krävs för att förbättra noggrannheten särskilt i de perifera delarna av lungorna.

Abstract [en]

Manual analysis of non-contrast computed tomography (CT) images, focusing on expiratory pulmonary blood vessels, remains a significant challenge. To facilitate and improve diagnosis in lung diseases such as interstitial lung diseases (ILD), including pulmonary fibrosis, there is an increasing need for quantitative and automated image analysis methods. The aim of this study was to develop and evaluate an AI model for vessel segmentation based on nnUNet, a self-configuring neural network for medical image analysis. A total of 33 non-contrast thoracic CT scans in the expiratory phase were retrieved from Karolinska University Hospital in Solna via the PACS database. The images were segmented in 3D Slicer and used to train the model with nnUNet through several iterative training rounds. The model’s performance was quantitatively evaluated using the Dice Similarity Coefficient (DSC), compared against manually segmented reference images from seven randomly selected cases. The results showed a gradual improvement in the model's ability to identify an increasing portion of the vascular tree, particularly in central lung regions. The final model, after three training rounds, achieved a mean DSC of 0.991. However, segmentation was less accurate in peripheral vessel regions. nnUNet shows promising results for automated segmentation of pulmonary vessels in expiratory, non-contrast CT images and has potential for future clinical applications. Further development and expansion of training data are needed to improve accuracy, especially in peripheral lung areas.

Place, publisher, year, edition, pages
2025. , p. 48
Series
TRITA-CBH-GRU ; 2025:166
Keywords [en]
Computed Tomography CT, Pulmonary vessels, 3D Slicer, nnUNet, Neural networks, Medical image analysis, Automatic segmentation, AI in radiology
Keywords [sv]
Datortomografi DT, Pulmonella blodkärl, 3D Slicer, nnUNet, Neurala nätverk, Medicinsk bildanalys, Automatisk segmentering, AI i radiologi
National Category
Clinical Medicine
Identifiers
URN: urn:nbn:se:kth:diva-365735OAI: oai:DiVA.org:kth-365735DiVA, id: diva2:1978315
External cooperation
Karolinska Universitetssjukhuset
Subject / course
Medical Engineering
Educational program
Master of Science in Engineering - Medical Engineering
Supervisors
Examiners
Available from: 2025-09-15 Created: 2025-06-27 Last updated: 2025-09-15Bibliographically approved

Open Access in DiVA

fulltext(2937 kB)61 downloads
File information
File name FULLTEXT01.pdfFile size 2937 kBChecksum SHA-512
595404b00031fabf7a1c26a9b0245c8dd17c93c7477be92d28ad7fa7983e084b5eaa00c5ec670b2006c377046ea2969e9e6c79a7a3a4b99dcc7f63a6559e7f8d
Type fulltextMimetype application/pdf
bilaga(87 kB)73 downloads
File information
File name ATTACHMENT01.txtFile size 87 kBChecksum SHA-512
c3a247df8f9d39fbbd3b57a3f4c69b8b15ad3a1601489c6d39f8e4cb7f15fd172fd24f8e7288a402c2d25622b4ba21a42ba586694771a503266ee5b5d84a314d
Type attachmentMimetype text/plain

By organisation
Biomedical Engineering and Health Systems
Clinical Medicine

Search outside of DiVA

GoogleGoogle Scholar
Total: 61 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 303 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf