Abstract: TH-PO573
Automated Tools for Renal Biopsy Diagnosis
Session Information
- Pathology and Lab Medicine
November 03, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
Abstract Time: 10:00 AM - 12:00 PM
Category: Pathology and Lab Medicine
- 1700 Pathology and Lab Medicine
Authors
- Li, Zhongwang, University College London, London, London, United Kingdom
- Walsh, Stephen B., University College London, London, London, United Kingdom
- Siew, Keith, University College London, London, London, United Kingdom
Background
Histological examination of the glomeruli in renal biopsies is essential for diagnosing many kidney diseases. However, manual identification and characterisation of glomeruli in biopsies requires trained histopathologists and can be tedious time-consuming task. Thus, development of automatic tools to help accelerate diagnostic workflow and improve detection accuracy is an unmet need. Such tools should have the ability to detect and count glomeruli of various sizes, shapes, and disease status across whole slide image (WSI) biopsies, annotate these in the image. Advanced functionality could then be developed to classify glomeruli based on disease features (e.g. fibrosis, mesangial expansion) and generate preliminary biopsy reports for review by the histopathologist.
Methods
Routine renal biopsies collected at Royal Free Hospital with patients’ consent for research were used for this study (n=300). 2-5um biopsy sections were acquired as 8-bit RGB WSI using an Axio Scan Z.1 (20x/0.8NA). To ensure the general applicability of the model, multiple common histochemical stains (H&E, PAS, Silver stain) were imaged and included in the training dataset. These digitised slides were then manually annotated by histotopathologists and nephrologists using QuPath (version 0.3.2) on Wacom Cintiq Pro 32 interfaces to capture the features described in the matching to biopsy reports.
“You only look once” (YOLO) is a real-time object detection system that has previously been used to identify glomeruli in PASM stained section. For our work, we decided to adapt this approach using the latest version of YOLO that can function under multiple conditions (e.g. different stains and magnifications), and a Convolutional Neural Network (CNN) model may then be used to classify various glomerular diseases.
Results
Using the location information given by YOLO, the glomerular images could be cropped from the original image and tabularised alongside morphometric and histolopathological readouts from our U-NET model that could segment the identified glomeruli from the background tissue and quantify features (e.g. % area of fibrosis).
Conclusion
This work shows that an automated image analysis pipeline can identify, quantify and characterize glomeruli in 2D slides in seconds, with obvious utility for pathologists and clinicians.
Further work to expand the functionality of this model and to validate it further in larger datasets is warranted.