Abstract: TH-PO017
Machine Learning Classification of Kidney Biopsy Smartphone Images for Adequacy Assessment
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania 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
- Cassol, Clarissa Araujo, Arkana Laboratories, Little Rock, Arkansas, United States
- Coley, Shana M., Arkana Laboratories, Little Rock, Arkansas, United States
- Kenan, Daniel J., Arkana Laboratories, Little Rock, Arkansas, United States
- Napier, Johnathan Oliver Howard, Arkana Laboratories, Little Rock, Arkansas, United States
- Sharma, Shree G., Arkana Laboratories, Little Rock, Arkansas, United States
Background
Kidney biopsy is the gold standard for diagnosis of medical renal diseases. A biopsy that yields predominantly medulla or not enough renal cortex is an unsatisfactory result. There has been a significant increase in the rate of inadequate kidney biopsies. Unfortunately, not all centers have access to trained professionals who can assess biopsy adequacy in real time. Therefore, we aim to create a machine-learning model capable of classifying smartphone images of kidney biopsy tissue as adequate or inadequate.
Methods
747 kidney biopsy cores and corresponding smartphone macro images were obtained from unused deceased donor kidneys. Each core was imaged, formalin fixed, sectioned and stained with Periodic acid–Schiff (PAS). A photo of the fresh unfixed core was taken using the macro camera on an iPhone 13 Pro. The amount of cortex in each core (percent cortex), was determined by two renal pathologists review of the PAS sections. Biopsies with less than 30% cortex were labelled as inadequate. Biopsies with 30% or more cortex were labeled as adequate. The images were split into a training (n=643), validation (n=30), and test (n=74) sets. The preprocessing steps were converting from HEIC iPhone format to JPEG, normalizing, and detecting the renal tissue; a U-Net deep learning model was trained to segment renal tissue from the background. After preprocessing, a deep learning model was trained on the renal tissue region of interest and corresponding class label. See Figure 1.
Results
The deep learning model had an accuracy of 87% on the training data. On the test dataset, the model had an accuracy of 82%. For inadequate samples in the test dataset, the model had a sensitivity of 71%. The area under the receiver operating curve was 0.79.
Conclusion
We developed and tested a machine learning model to classify smartphone images of kidney biopsy as adequate or inadequate, based on the amount of cortex determined by a renal pathologist. With further work, such models can be deployed as a smartphone application to aid in real time assessment of renal biopsy adequacy.
Figure 1. Methodology Overview
Funding
- Other NIH Support