Abstract: TH-PO027
A Deep-Learning Approach to Grading Kidney Interstitial Fibrosis: Evaluation of a VGG-16 Model on Trichrome-Stained Biopsy Images
Session Information
- Augmented Intelligence for Prediction and Image Analysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Eigbire-Molen, Odianosen J., Arkana Laboratories, Little Rock, Arkansas, United States
- Muni Reddy, Swathi, University of Arkansas for Medical Sciences, Little Rock, Arkansas, United States
- Sharma, Shree G., Arkana Laboratories, Little Rock, Arkansas, United States
Background
Renal interstitial fibrosis is an important morphologic predictor of outcome in kidney disease. Grading of interstitial fibrosis is routinely performed by renal pathologists during renal biopsy evaluation. The most common technique for grading interstitial fibrosis is visual estimation on stained sections, typically a trichrome stain. However, this technique is prone to interobserver variability. Deep learning models have shown good performance in image classification tasks with robust reproducibility. We present a deep learning classification model trained to grade renal interstitial fibrosis using static low magnification images of trichrome stained sections.
Methods
505 low magnification overview images of trichrome stained renal biopsy sections were obtained from the archive of cases signed out in 2023 at Arkana Laboratories. A VGG-16 deep learning model was trained to classify the images into four categories - no fibrosis (0-5%), mild fibrosis (6-25%), moderate fibrosis (26-50%), and severe fibrosis (51-100%). The gold standard was the signout pathologist grading. The dataset was split into a train (80%), validation (10%) and test set (10%). The model consisted of a base VGG-16 model with a dense layer and softmax classifier for categorical classification (Figure 1).
Results
The deep learning model attained an overall accuracy of 95% on the training dataset and 92% on the validation dataset. In comparison to the pathologist gold standard, the deep learning model had an overall accuracy of 82% and differentiated between varying degrees of fibrosis (Figure 2).
Conclusion
This study demonstrates the application of a deep learning model to reproduce grading of renal interstitial fibrosis. The model was able to differentiate various grades of interstitial fibrosis, demonstrating a potential prognostic value.
Figure 1: Model
Figure 2: Model Accuracy by Fibrosis