Medizinische Universität Graz Austria/Österreich - Forschungsportal - Medical University of Graz
Gewählte Publikation:
Murauer, C.
HUMAN VS. MACHINE: PD-L1 IMMUNOHISTOCHEMISTRY EVALUATION IN NON-SMALL CELL LUNG CARCINOMA AND PLEURAL MESOTHELIOMA
Humanmedizin; [ Diplomarbeit ] Medizinische Universität Graz; 2021. pp. 65
[OPEN ACCESS]
FullText
- Autor*innen der Med Uni Graz:
- Betreuer*innen:
- Brcic Luka
- Zacharias Martin
- Altmetrics:
- Abstract:
- Introduction: Digitalization, the use of artificial intelligence (AI), and machine learning (ML) are increasingly used to support pathologists in diagnostic procedures. Such technologies aim to achieve faster, more objective, and more accurate results compared to human evaluation. In many areas of pathology, AI/ML is already used as a supporting tool in routine work and provides promising results. PD-L1 scoring in lung carcinomas is very subjective and varies among pathologists. However, it is very important for the decision about immunotherapy. This study aimed to compare the PD-L1 scoring of lung carcinomas and mesotheliomas between AI and humans. Methods: In this retrospective study, 51 samples of NSCLC, diagnosed in 2020 at the Diagnostic and Research Institute of Pathology at the Medical University of Graz, and 24 samples of pleural mesothelioma, diagnosed between 2018 and 2020, were included. All 75 samples were divided into three groups: small biopsy samples (NSCLC), resected tumors (NSCLC), and mesothelioma samples. uPath PD-L1 (SP263) image analysis by Ventana (Roche) was used for automated scoring. Those results were then compared to the evaluated scores of an experienced thoracic pathologist and a student who analyzed the samples under the microscope. Results: Concordance between the AI scores and the pathologist/student scores varied between the three groups. In the small biopsy group, the calculated Cohen’s kappa score showed a slight agreement. The results for both resected samples and mesothelioma samples were classified as poor agreement. Conclusion: This study shows that the concordance between AI scoring and human scoring among NSCLC and mesotheliomas is not as high as expected. However, our study has several limitations, and further studies are needed to confirm or dispute our preliminary results.