Get started
Work

Automatic segmentation of epithelium in cervical whole slide images to assist diagnosis of cervical intraepithelial neoplasia

Year: 2020
Type: dissertation
Abstract: Cervical Intraepithelial Neoplasia (CIN) is a precancerous state of the cervix, and the correct \ndiagnosis of its level of severity (CIN grade) allows to determine patient treatment and prevent \ninvasive carcinoma. It is diagnosed from histological samples by visually estimating the \npercentage of epithelial width covered by neoplastic cells, although the subjective nature of this \nestimation has proved to cause high levels of inter-observer variability in the diagnosis. \nImmunohistochemical (IHC) stains have been used to increase inter-observer agreement, and to \nperform immune cell analysis for cervical cancer research, but the manual quantification of IHC \nis still prone to subjective interpretation, and unfeasible to implement on a large scale. \nDigital pathology enables the digitalization of samples into Whole Slide Images (WSI), using \ntissue scanners that apply the principles of light microscopy. Because CIN occurs in the \nepithelial tissue, the first step towards a computer-assisted quantification of IHC stains would \nbe to automatically segment the epithelium in WSIs. Deep Learning (DL) has been used in \ndigital pathology to detect histological regions with lesions, quantify biomarkers and segment \ntissue types. \nIn digital pathology, image properties such as color, brightness, contrast or blurriness may vary \nbased on the scanner, the tissue preparation and the technician handling the equipment. The \napplication of DL for image analysis in this field must be robust to image variability, and deliver \nconsistent results regardless of the origin or handling of the image. Stability Training (ST) was \nevaluated as a potential method to increase robustness of DL models in digital pathology. During \nmodel training, ST generates a distorted version of the training images, and penalizes differences \nbetween the prediction of the original image and the prediction of the distorted version. \nThis work evaluated if the application of ST could improve the DL models’ robustness to image \nvariations caused by the use of different IHC stains and different scanners, and the results \nobtained suggest that it makes DL models better at handling WSIs with image variations. \nThree scanner models (NanoZoomer-HT, NanoZoomer-XR and NanoZoomer-S360) and three \nIHC stains (p16, CD3 and CD8) were used for this work. 114 WSIs stained with the p16 IHC and scanned with the NanoZoomer-HT (p16-HT slides) were used to train a set of DL models \nwith different weights for the stability component (α) and different distortion combinations, \nincluding a model with no ST for reference (α=0). The models were tested on six test sets (23- \n24 WSIs each) with different IHC stain and scanner combinations: p16-HT, p16-XR, p16-S360, \nCD3-HT and CD8-HT. An additional test set was used to evaluate robustness to blurriness \ncaused by lack of lens focus (p16-HT-out of focus). Area Under the ROC Curve (AUC) was \nused to evaluate model performance. \nThe models with ST outperformed the model without ST in all test sets. All models had a similar \nperformance on the p16-HT test set (AUC of 0.978-0.985). For p16-HT-out of focus, the model \nwith no ST had a significantly lower performance than the best model with ST (AUC 0.876 vs \n0.939). The same was observed for p16-XR (AUC 0.913 vs 0.962), p16-S360 (AUC 0.840 vs \n0.962), CD3-HT (AUC 0.973 vs 0.985) and CD8-HT (AUC 0.961 vs 0.980). Distortion \ncombination 3 (high image distortion) and α=10 were provided the best ST results.
Language: English
Open Access status: green
OpenAlex ID: W3214963592
Cites: 60
Cited by:
Related to: 10