ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: TH-PO575

Development of a Computational Modeling Tool for Automated Detection of Urinary Casts and Acanthocytes

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

  • Flores Santiago, Josean O., Ochsner Health Network LLC, New Orleans, Louisiana, United States
  • Velez, Juan Carlos Q., Ochsner Health Network LLC, New Orleans, Louisiana, United States

Group or Team Name

  • Ochsner Nephrology
Background

Microscopic examination of the urinary sediment (MicrExUrSed) is a tool of proven clinical utility. However, it requires training and expertise for proper identification of urinary casts. Incorrect cast identification my lead to flawed clinical-decision making. We hypothesized that a machine learning approach could be used to create a tool for automated real-time identification of urinary casts.

Methods

We accessed a database of MicrExUrSed images routinely obtained and stored as part of a prospective research cohort. Images were captured with a smartphone camera adapted to the microscope eyepiece. Images were categorized and labeled by trained observers and divided into datasets for classification and object detection tasks. Input contained annotations about illumination, staining, and magnification. Bounding boxes for 3 major structures: casts, cells, and crystals were fed into the model. We performed preliminary experimentation on a subset of images (308 images containing 572 casts and 107 acanthocytes) to train a YOLOv5 object detection model to identify acanthocytes and 4 types of clinically relevant urinary casts: muddy brown-granular, granular, waxy casts, and cellular casts. Images were preprocessed using YOLOv5 data augmentation technique. The model was trained with empty weights and on pre-trained models, with image sizes of 640 and 1280, batches of 6 and 16, and 300 epochs.

Results

After model development and training, performance was tested in 75 additional images containing acanthocytes and casts. The best performance was obtained with YOLOv5m with 640 image size, a batch of 6, and 300 epochs. It achieved a mean average precision (mAP) of 0.7618, which is above the performance of benchmark databases. The model performed equally accurately for acanthocytes and casts.

Conclusion

To our knowledge, this is the first implementation of computational model/machine learning for real-time automated identification of findings from MicrExUrSed. In this dataset, the different illumination techniques, degree of magnification, high-resolution images, and staining techniques served as different backgrounds that increased the model performance even with a small number of training data. More work is needed to expand the capability of the tool to other unique structures.