Artificial Intelligence in clinical diagnostic pathology: a comparison of image analysis methods to detect Ki67 positive cells in breast cancer
Virtanen, Paavo (2024)
2024
Bioteknologian ja biolääketieteen tekniikan maisteriohjelma - Master's Programme in Biotechnology and Biomedical Engineering
Lääketieteen ja terveysteknologian tiedekunta - Faculty of Medicine and Health Technology
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-05-21
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202404254592
https://urn.fi/URN:NBN:fi:tuni-202404254592
Tiivistelmä
Breast cancer is a prevalent malignancy among women worldwide, posing a significant health burden for both patients and healthcare, with millions of new cases diagnosed annually around the world. While predominantly affecting women, a small percentage of cases also occur in men. Accurate assessment of cellular proliferation, crucial for determining cancer grade and prognosis, relies on markers such as the Ki67 protein. The evolution of Ki67 index calculation, from manual counting to artificial intelligence (AI)-based approaches, reflects the broader transition of pathology towards digitalisation. Digital pathology offers numerous advantages, including enhanced efficiency, diagnostic accuracy, and accessibility for remote consultation. AI holds promise for further improving diagnostic processes by providing tools for quick identification of regions of pathological interest and reducing interobserver variability between pathologists, making diagnostics more consistent.
Fimlab, Finland's leading laboratory company, sought a new software solution for Ki67 index calculation following the retirement of their current software SlideVantage. Aiforia provides AI tools for Ki67 index analysis, and its AI model trained on breast cancer tissue sections is one of the possible successors. However, to meet Fimlab's requirements, Aiforia had to develop a retrained pilot model compatible with the Philips Ultra Fast Scanner's whole slide images (WSIs).
After providing an extensive background on relevant themes, this thesis evaluates the performance of Aiforia's browser-based deep learning Ki67 index analysis software in comparison to SlideVantage. The assessment primarily focuses on the performance of cell recognition, classification, and index calculation capabilities of both platforms. In addition, the overall performance of Aiforia’s software was tested. Aiforia's pilot model, coupled with Philips Ultra Fast Scanner, demonstrated superior cell detection capabilities verified by an experienced pathologist, offering potential for more accurate Ki67 index analysis in near future. Despite minor issues such as occasional inconsistencies in heatmap, the software shows promise as a possible successor to SlideVantage. This study underscores the importance and opportunities of digital pathology and AI in enhancing future cancer diagnostics and treatment decision-making.
Fimlab, Finland's leading laboratory company, sought a new software solution for Ki67 index calculation following the retirement of their current software SlideVantage. Aiforia provides AI tools for Ki67 index analysis, and its AI model trained on breast cancer tissue sections is one of the possible successors. However, to meet Fimlab's requirements, Aiforia had to develop a retrained pilot model compatible with the Philips Ultra Fast Scanner's whole slide images (WSIs).
After providing an extensive background on relevant themes, this thesis evaluates the performance of Aiforia's browser-based deep learning Ki67 index analysis software in comparison to SlideVantage. The assessment primarily focuses on the performance of cell recognition, classification, and index calculation capabilities of both platforms. In addition, the overall performance of Aiforia’s software was tested. Aiforia's pilot model, coupled with Philips Ultra Fast Scanner, demonstrated superior cell detection capabilities verified by an experienced pathologist, offering potential for more accurate Ki67 index analysis in near future. Despite minor issues such as occasional inconsistencies in heatmap, the software shows promise as a possible successor to SlideVantage. This study underscores the importance and opportunities of digital pathology and AI in enhancing future cancer diagnostics and treatment decision-making.