MobileNetV2-Based Deep Learning Model for Efficient Classification of Kidney Disorders

Abstract

Kidney stone also known as renal calculi, is a painful urological condition which can lead to serious complications if not detected timely. In the present times, there has been a surge in the number of kidney stone patients. Conventional approaches have several drawbacks in detection of kidney diseases including delays, incorrect predictions and a lot of manual effort. Deep Learning (DL) based algorithms have overcome these limitations and offered a promising solution called Convolution Neural Networks (CNN) which is excellent in capturing the complex image patterns. CNNs have transformed the task of medical imaging by enabling more accurate and efficient predictions. In this study kidney image classification is done using MobileNetV2, a lightweight, pretrained CNN architecture and the images are categorized into four classes that are cyst, stone, tumor and normal. The model demonstrated excellent outcomes by achieving an accuracy of 96 %, recall of 95% and precision of 96%. The results of this study proves the capability of CNNs in clinical tasks for enhancing patient care.

Introduction

Kidneys play a vital role in maintaining the body’s internal balance by filtering blood, removing waste, and regulating blood pressure. Kidney stones—formed from the buildup of mineral and salt crystals—can cause severe pain and urinary infections, making early detection essential for proper treatment. Traditional diagnostic techniques such as CT scans, ultrasounds, and X-rays, though widely used, can be time-consuming and prone to human error.

Deep Learning (DL), especially Convolutional Neural Networks (CNNs), has recently emerged as a powerful solution for automatic kidney stone detection. CNNs excel at extracting features from medical images and improving classification accuracy. This study focuses on the use of MobileNetV2, a lightweight pretrained CNN model, to accurately classify kidney images and assist healthcare professionals in early diagnosis and better patient management.

Literature Review

Various studies have used hybrid CNN–SVM models, Capsule Networks, Mask-RCNN, EfficientNet, YOLO-based detectors, and ensemble learning to improve kidney stone detection. These methods aim to enhance accuracy, handle class imbalance, and address spatial localization, but challenges remain, such as high computation costs, small datasets, incomplete dataset information, and reduced performance on imbalanced data.

Materials and Methods

• Dataset: A Kaggle CT kidney image dataset with 12,446 images across four classes—Normal, Cyst, Stone, Tumor.

• Model: MobileNetV2 (lightweight, fast, suitable for transfer learning).

• Techniques Used: Transfer learning, image preprocessing (median blur, CLAHE), data augmentation, normalization, and resizing images to 224×224.

• Evaluation Metrics: Accuracy, Precision, Recall, and F1-Score.

Proposed Methodology

The dataset is split into training/validation/testing (70/15/15). Images are preprocessed and augmented to improve model generalization. MobileNetV2’s early layers are frozen and the top 40 layers are fine-tuned for kidney image classification. Global Average Pooling and Dropout are used to reduce overfitting, and the model is trained for 30 epochs with callbacks like EarlyStopping and ReduceLROnPlateau.

Results and Discussion

The model shows excellent performance:

• Training Accuracy: 96.59%

• Validation Accuracy: 97.05%

• Testing Accuracy: 96%

• F1-Score: 95%

High precision and recall show strong reliability in distinguishing diseased and healthy images. Training and validation curves indicate stable learning without overfitting.

Conclusion

This paper presents the use of MobileNetV2 architecture in multiclass classification of Kidney CT images. The results of this study shows that a lightweight model can also give promising outcomes if finetuned carefully. It also shows that automated detection methods, aided by deep learning can decrease the probability of misdiagnosis and can enhance the quality of treatment plans. Future studies can focus on utilizing ensemble methods, explainable AI, training on complex and diverse datasets and integrating the model into real time clinical applications.

References

[1] L. S. Rapelang and I. C. Obagbuwa, “Hybrid Support Vector Machine–Convolutional Neural Networks Multi-Classification Models for Detection of Kidney Stones,” International Journal of Imaging Systems and Technology, vol. 35, no. 4, e70128, 2025, doi: 10.1002/ima.70128. [2] A. K. Shukla, P. Anusha, G. B. S. R. Naidu, G. Sabarinathan, N. L. Kumar, and M. Ponnusamy, “An image processing method for kidney stone segmentation in CT scan images based on CNN-regularized extreme learning machine approach,” in Hybrid and Advanced Technologies, vol. 1, S. P. J. Christydass, N. Nurhayati, and S. Kannadhasan, Eds. Taylor & Francis, 2025, pp. 217–222, doi: 10.1201/9781003559115-37. [3] O. H. Jasim, D. A. Q. Shakir, M. S. Hamad, and W. K. Awad, “Precise kidney stone localization in medical imaging via a capsule network,” Mesopotamian Journal of Big Data, vol. 2025, pp. 136–143, Aug. 2025, doi: 10.58496/MJBD/2025/009. [4] F. Alam, K. Alam, and N. Khan, “An automatic approach for the detection and segmentation of kidney stone in KUB CT images using Mask R-CNN,” Technical Journal, University of Engineering and Technology (UET) Taxila, vol. 30, no. 1, pp. 47–56, 2025. [5] V. Rathva and S. Degadwala, “Ensemble Deep Learning for Enhanced CT-Scan Kidney Stone Classification,” International Journal of Scientific Research in Science and Technology, vol. 12, no. 3, pp. 432–439, May 2025, doi: 10.32628/IJSRST2512359. [6] S. Bala, K. Arora, V. Satheeswaran, S. Mohan, D. J., K. Sangamithrai, and A. N. Doss, “Improving Kidney Stone Detection with YOLOV10 and Channel Attention Mechanisms in Medical Imaging,” Journal of Electronics, Electromedical Engineering, and Medical Informatics, vol. 7, no. 3, pp. 951–963, Jul. 2025, doi: 10.35882/jeeemi.v7i3.868. [7] K. Mohammed, M. Sravani, B. D. Manimala, M. K. Raza, and A. T. Vishal, “Kidney Stone Detection by Using Intelligent Segmentation with Ensemble Deep Neural Network,” International Journal of Advances in Applied Sciences and Engineering, vol. 4, no. 1, pp. 14–22, Mar. 2025. [8] N. Panchal, M. M. Raikar, and V. P. Baligar, “Kidney stone detection using deep learning model,” Procedia Computer Science, vol. 260, pp. 56–63, 2025, doi: 10.1016/j.procs.2025.03.177. [9] S. Yalç?n, \"Kidney stone detection using an EfficientNet-based method,\" Journal of Computer Science (Anatolian Science), vol. 10, no. 1, pp. 1–10, 2025, doi: 10.53070/bbd.1623346. [10] D. S. and P. Valarmathi, \"Deep learning-based detection of kidney stones using VGG16 and transfer learning techniques,\" Journal of Multidimensional Research and Review (JMRR), vol. 6, no. 1, pp. 154–161, Apr. 2025.

Copyright

Copyright © 2025 Gauri Mittal. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.