Taha, Anastasia (2026) From eyeballing to deep learning: the comparison of manually and automated methods to determine the Ki-67 proliferation index in gastroenteropancreatic neuroendocrine neoplasms. thesis, Medicine.
Full text available on request.Abstract
NENs (Neuroendocrine neoplasms) are rare tumours, often located in the digestive tract. The tumour grade classification is determined by the Ki-67 proliferation index (Ki-67 PI), which determines the prognosis and treatment. The golden standard (GS) for this is manual counting, which is prone to inter- and intra-observer variability, indicating that there is a need for a more standardised method. The aim of this study is to compare the agreement for two manual methods: eyeballing and manual counting and three deep learning-based methods: virtual double staining, manual hotspot selection and automatic hotspot selection, for determining the Ki-67 PI and the tumour grade classification in 47 gastroenteropancreatic NENs. The agreement between each method with the GS was determined with Fleiss’s and weighted Cohen’s kappa for tumour grades and with the interclass correlation coefficient (ICC) for the Ki-67 PI. All methods had substantial to almost perfect agreement with the GS for tumour grading and excellent agreement for the Ki-67 PI (kappa=0.786-0.964; ICC=0.907-0.969). Deep learning-based methods consistently showed excellent agreement for the Ki-67 PI and almost perfect agreement for tumour grading. Among the deep learning-based methods, manual hotspot selection had the highest agreement with the GS for tumour grading and the Ki-67 PI (kappa =0.928; ICC= 0.967). From the manual methods, eyeballing showed the greatest interobserver variability (kappa: 0.786-0.906; ICC: 0.907-0.965). This indicates that deep learning-based algorithms could be used in the future to determine the Ki-67 PI more consistently, leading to a more accurate prognosis and treatment.
| Item Type: | Thesis (UNSPECIFIED) |
|---|---|
| Supervisor name: | Kats-Ugurlu, Dr. Gursah |
| Faculty: | Medical Sciences |
| Date Deposited: | 14 Oct 2026 13:33 |
| Last Modified: | 14 Oct 2026 13:33 |
| URI: | https://umcg.studenttheses.ub.rug.nl/id/eprint/3965 |
Actions (login required)
 |
View Item |