ADVANCING KIDNEY DISEASE DIAGNOSIS : A MACHINE LEARNING APPROACH FOR MULTICLASS CLASSIFICATION USING COMPUTED TOMOGRAPHY (CT) DATA

Introduction:

Kidney diseases, including cysts, stones, and tumors, are major health concerns worldwide, affecting millions of individuals and posing significant challenges to healthcare systems . Accurate and timely diagnosis is crucial for effective treatment and management. Traditional diagnostic methods often rely on manual interpretation of CT scans, which can be time-consuming and prone to human error. This paper proposes an advanced machine learning (ML) model for the classification of kidney diseases using CT data. By leveraging a dataset of 12,400 kidney images, the proposed model aims to efficiently classify four disease classes: normal, cyst, stone, and tumor. The investigation evaluates the performance of pretrained MobileNetV2 model, in terms of accuracy. The results demonstrate that the ML model can significantly enhance the accuracy and speed of kidney disease diagnosis

Methods:

The dataset of 12,400 kidney CT images (taken from Kaggle Dataset) was organized into four classes: normal, cyst, stone, and tumor. Images were resized to 224x224 pixels, normalized to a [0, 1] range, and augmented using techniques like rotation and flipping. The dataset was split into training (70%), validation (15%), and test (15%) sets, ensuring well-prepared data for enhancing the model's accuracy and robustness in kidney disease classification.

MobileNetV2 was selected for this project due to its balance between computational efficiency and accuracy, making it ideal for real-time clinical use . Its use of depth-wise separable convolutions reduces parameters, allowing high accuracy with minimal resources. The model was trained using categorical cross-entropy loss and the Adam optimizer, with regularization techniques like dropout and early stopping to prevent overfitting. Implemented in TensorFlow and Keras, the model was trained on a Dell workstation to expedite processing. A user-friendly Gradio interface was developed for clinicians to easily upload and classify CT images, providing real-time diagnostic feedback.

Results:

Figure 1 shows the model's performance during training and validation. Training accuracy steadily improves, and validation accuracy trends upward, indicating good generalization. Training and validation losses both decrease, reflecting fewer errors and strong performance. These trends suggest the model is effective in learning and classifying kidney diseases, making it a robust tool for kidney disease classification task.

Figure 2 depicts the interface of a Jupyter Notebook designed for a “Kidney Disease Prediction” application. The interface includes buttons for image upload, input clearing, and submission for analysis. Notably, the cell below the interface is currently empty, indicating that no output has been generated at this stage. From Figure 3 (a-c), this tool uses MobileNetV2 for efficient classification of kidney diseases, including normal, cyst, stone, and tumor conditions. It features a user-friendly interface where CT scan images are analyzed, with results displayed after submission.

Conclusions:

The high accuracy and efficiency of the MobileNetV2 model suggest that it can be a valuable tool in clinical practice, assisting radiologists in the rapid and accurate diagnosis of kidney diseases. This automated approach can help to reduce diagnostic errors and improve patient outcomes by providing consistent and reliable classifications, especially in high-volume settings where time and accuracy are critical. The efficiency of this model can be improved by increasing number of epochs during training and also unfreezing the underlying layers during model training.

I have no potential conflict of interest to disclose.

I did not use generative AI and AI-assisted technologies in the writing process.