Automated Urine Sediment Microscopic Image Analysis Using YOLOv10 with Chimpanzee Leader Election Optimization

Summary

The study proposes a model utilizing the UMID dataset to detect and classify urine cells, including Red Blood Cells, White Blood Cells, epithelial cells, crystals, and bacteria. Employing YOLOv10 with Chimpanzee Leader Election Optimization, it attains 84% mAP50 on a test subset, potentially decreasing manual microscopy workload in clinical urinalysis.

Abstract

To diagnose various urinary diseases, it is vital to accurately detect urine cells. This paper presents a novel approach for detecting and classifying urine cells using the YOLOv10 (You Only Look Once version 10) model integrated with the CLEO algorithm. YOLOv10, known for its superior performance in object detection, is combined with CLEO, a nature-inspired optimization algorithm. The model is trained on the Urine Microscopic Image Dataset (UMID), comprising various types of urine cell images. Multiple cell types have been classified, including bacteria, epithelial cells, RBCs, WBCs, crystals, and casts directly from microscopic images of urine samples. Results from rigorous testing show that YOLOv10 integrated with CLEO is a robust tool for analyzing urine sediments, achieving a mean Average Precision@50 (mAP50) of 84% and a recall of 90%. Hence, the proposed approach offers significant speed improvements and consistency as compared to traditional methods of urine sediment analysis. The framework can automatically analyze urine sediment images in real time with good detection performance and has the potential to assist computer-assisted clinical urinalysis, thereby reducing the workload of manual analysis and diagnosis.

Introduction

Kidney disease, urinary tract infections, metabolic disorders, and systemic abnormalities can be detected by urinalysis, which is a very important process of diagnosis^1,2. It is essential to examine the urine sediments under the microscope to identify miniature constituents such as Red Blood Cells (RBC), White Blood Cells (WBC), bacteria, fungi, casts, etc³. However, conventional manual analysis by optical microscopy is labor-intensive and depends on the skill of the technician and may lead to inter-observer variability and inconsistent diagnoses in busy clinical environments

Protocol

The data used in this study came from a public research database of microscopic urine sediment images. During the study, no personal information identifying the patient was used. The commercially available elements associated with this protocol are listed in the Table of Materials.

Acquire and Prepare Dataset
Microscopic urine sediment images comprising RBCs, WBCs, epithelial cells, crystals, casts, bacteria, and yeast were obtained. All images were saved in PNG or JPEG format using consistent naming conventions. Blurry or poor-quality images containing incomplete sediment artifacts were eliminated. Bou....

Results

The training and validation loss curves are shown in Figure 4, which depicts the YOLOv10 combined with CLEO performance metrics for urine cell detection. The accuracy of the predicted bounding boxes for detecting urine cells is measured by the box loss. The error in the prediction of the correct class is measured by the classification (cls) loss. The distributed focal loss (dfl) is used to improve bounding box regression accuracy. The values of all the training curves vary from 1.0 to approx.......

Discussion

The results of this study provide compelling evidence of the potential of automated detection and classification of urine sediment components with the help of YOLOv10 and CLEO in the context of medical diagnostics. The proposed model provides a faster and more reliable solution for making timely decisions by optimizing YOLOv10 hyperparameters with CLEO. The proposed system provides high precision, recall, F1-score, and mAP values for various urine cells such as RBCs, WBCs, epithelial cells, bacteria, crystals and casts.<.......

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
CUDA Toolkit 12.1	NVIDIA Corporation	Version 12.1/ RRID:SCR_018425	GPU acceleration platform
LabelImg Annotation Tool	Tzutalin	RRID:SCR_016189	Annotation tool
Matplotlib	Matplotlib Development Team	Version 3.x	Data visualization
NumPy	NumPy Developers	Version 1.x	Numerical computation
OpenCV	OpenCV.org	RRID:SCR_015881	Image pre-processing and analysis
Pandas	Pandas Development Team	Version 2.x	Data handling and analysis
Python 3.10	Python Software Foundation	RRID:SCR_008394	Programming language
PyTorch	PyTorch Foundation	RRID:SCR_018536	Deep learning framework
Scikit-learn	scikit-learn Developers	RRID:SCR_002577	Machine learning utilities
Ubuntu Linux	Canonical Ltd.	Version 22.04	Operating system
Urine Sediment Microscopic Dataset	Public dataset repository	GitHub - dipamgoswami/UMID-Urine-Microscopic-Image-Dataset: Urine Microscopic Image Dataset with point and box annotations · GitHub	Urine microscopy image dataset
YOLOv10 implementation	Ultralytics	YOLOv10: Real-Time End-to-End Object Detection | Ultralytics Docs	Object detection framework