UroVision: AI-Based Kidney Stone Detection System

UroVision is an advanced deep learning-based system designed to detect kidney stones from medical images. Built using the YOLOv8n object detection model and deployed with Streamlit, this project aims to provide an efficient, real-time diagnostic tool for healthcare professionals.

Wanna try the model ? https://urovision.streamlit.app/

---

Table of Contents

• Introduction
• Features
• Dataset
• Model Architecture
• Installation
• Usage
• Results
• Challenges and Future Work
• Contributors

---

Introduction

Kidney stone diagnosis often relies on manual examination of medical imaging, which can be time-consuming and error-prone. UroVision automates this process using state-of-the-art AI techniques, offering:

• High detection accuracy.
• Real-time prediction via an interactive web interface.
• A scalable, easy-to-deploy solution.

---

Features

• Efficient Object Detection: Utilizes YOLOv8n, a lightweight yet powerful object detection model.
• Interactive Deployment: Built with Streamlit, allowing easy image uploads and real-time visualization of results.
• Scalability: Model optimized for deployment in clinical or cloud environments.

---

Dataset

The model is trained on a dataset sourced from Roboflow, consisting of annotated images of kidney stones.

• Training Data: 80% of the dataset.
• Validation Data: 20% of the dataset.

---

Model Architecture

YOLOv8n is chosen for its balance between performance and computational efficiency. The model was trained with the following parameters:

• Epochs: 32
• Batch Size: 32
• Learning Rate: 0.001
• Framework: Ultralytics (YOLO)

Trained weights are included in the weights directory for replication or deployment.

---

Installation

1. Clone the repository:

   git clone https://github.com/mohitmahajan095/UroVision.git
   cd UroVision
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Run the Streamlit app:

   streamlit run app.py
   

---

Usage

1. Launch the Streamlit app.
2. Upload an MRI image of Kidney via the interface.
3. View the detected kidney stones with bounding boxes and confidence scores.

---

Results

The system achieved the following metrics:

Precision: 84.4%

Recall: 70.3%

F1-Score: 0.77 at 0.378

1) Results

2) Precision-Recall Curve

3) F1-Score Curve

4) Confusion Matrix

---

Visual outputs:

Sample Predictions:

1) val_labels:

2) val_prediction:

---

Challenges

Challenges:

1. Limited dataset size.
2. Handling edge cases with complex imaging artifacts.

---

Contributors

1. Mohit Mahajan (Team Leader)
2. Mahadev Ambre
3. Kalpesh Patil
4. Ishwari Bhosale
5. Gauri Gawali
6. Shweta Shelke