Frederik A. Stuebs

Background: The purpose of this research is to estimate the rate of concordance, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of colposcopy for high-grade squamous lesions and carcinomas (HSIL+). Methods: We conducted a retrospective study of colposcopies performed in the certified Dysplasia Unit in Erlangen between January 2015 and May 2022 (7.5 years). The colposcopic findings were correlated with biopsies obtained during examinations or surgery. Cases without histology were excluded. The primary outcome was the rate of concordance between the colposcopic and histological findings in relation to the type of transformation zone (TZ), examiner’s level of experience and age of the patients. Results: A total of 4778 colposcopies in 4001 women were analyzed. The rates of concordance for CIN I/LSIL, CIN II/HSIL, CIN III/HSIL, and carcinoma were 43.4%, 59.5%, 78.5%, and 53.9%, respectively. The rate of concordance was lowest for TZ3 and highest for colposcopists with more than 10 years’ experience. Conclusions: Colposcopy is an important, feasible, and effective method. Careful work-up needs to be performed for women with TZ3 who are over 35 years old, as they are at the highest risk of being misdiagnosed. The highest concordance for detecting HSIL+ was seen for colposcopists with >10 years’ experience.

The tumor-stroma ratio (TSR) has been repeatedly shown to be a prognostic factor for survival prediction of different cancer types. However, an objective and reliable determination of the tumor-stroma ratio remains challenging. We present an easily adaptable deep learning model for accurately segmenting tumor regions in hematoxylin and eosin (H&E)-stained whole slide images (WSIs) of colon cancer patients into five distinct classes (tumor, stroma, necrosis, mucus, and background). The tumor-stroma ratio can be determined in the presence of necrotic or mucinous areas. We employ a few-shot model, eventually aiming for the easy adaptability of our approach to related segmentation tasks or other primaries, and compare the results to a well-established state-of-the art approach (U-Net). Both models achieve similar results with an overall accuracy of 86.5% and 86.7%, respectively, indicating that the adaptability does not lead to a significant decrease in accuracy. Moreover, we comprehensively compare with TSR estimates of human observers and examine in detail discrepancies and inter-rater reliability. Adding a second survey for segmentation quality on top of a first survey for TSR estimation, we found that TSR estimations of human observers are not as reliable a ground truth as previously thought.