AI-Based Kidney Stone Detection Using Ultrasound Imaging

Abstract

Kidney stone disease is a common urological condition where delayed detection can lead to severe complications. Diagnosis using ultrasound imaging depends on expert interpretation, which may not always be available in resource-limited settings. Manual analysis is time-consuming and pronetovariability,reducingthereliabilityofearly screening. Although artificial intelligence has shownpotentialinmedicalimaging,manyexisting systems lack robustness in handling irrelevant inputs. To address this, this work presents an AI-based kidney stone detection system using ultrasound imaging. The framework utilizes a MobileNetV2-based convolutional neural network with transfer learning to classify images into Normal and Kidney Stone Detected categories. A deep feature-based validation mechanism ensures that only relevant kidney ultrasound images are processed, preventing incorrect predictions. Grad-CAMvisualizationsareusedtohighlightimportant regions influencing the model’s decision. Experimental results demonstrate that the system provides accurate and reliable predictions. The model is deployed as a Flask-based web application, enabling real-time analysis with confidence scores and visual explanations. This work highlights the effectiveness of combining deep learning and explainable AI for kidney stone screening

Introduction

The text presents KidneyAI, a deep learning–based system for the automatic detection of kidney stones from ultrasound images. Kidney stone disease is a common urological condition that requires early diagnosis to prevent complications. While ultrasound imaging is affordable and non-invasive, its interpretation depends heavily on expert radiologists and can be challenging in resource-limited settings.

The study reviews previous research on kidney stone detection using machine learning and deep learning techniques. Earlier approaches using traditional image processing, handcrafted features, and machine learning methods achieved moderate accuracy but struggled with image variability and noise. Recent CNN and transfer-learning models improved performance, but many lacked mechanisms to verify whether uploaded images were actually kidney ultrasound images, leading to unreliable predictions.

To address these limitations, the proposed KidneyAI system uses a MobileNetV2-based transfer learning model for binary classification of ultrasound images into Normal and Kidney Stone Detected categories. The system includes an image validation module that detects and rejects irrelevant or non-medical images before classification, improving reliability. Additionally, Grad-CAM visualization is used to highlight image regions influencing the model’s decisions, increasing transparency and interpretability.

The methodology consists of data collection, image preprocessing, model training, evaluation, validation, visualization, and deployment. Images are resized to 224×224 pixels, normalized, and augmented before training. The model is trained using binary cross-entropy loss and the Adam optimizer, with regularization techniques such as dropout and data augmentation to reduce overfitting.

The system is deployed as a Flask-based web application where users can upload ultrasound images and receive real-time predictions, confidence scores, and visual explanations. The architecture includes modules for image upload, preprocessing, validation, classification, and result visualization.

Experimental results show that the MobileNetV2 model successfully learns complex ultrasound image patterns and provides accurate kidney stone detection. The image validation layer effectively prevents incorrect predictions on non-kidney images, while Grad-CAM heatmaps confirm that the model focuses on clinically relevant regions. Overall, the combination of deep learning, validation mechanisms, and explainable AI improves accuracy, reliability, robustness, and interpretability, making KidneyAI a practical and scalable solution for supporting early kidney stone diagnosis and assisting healthcare professionals.

Conclusion

The KidneyAI system successfully utilizes deep learning techniques to address the challenges of accurate and reliable kidney stone detection using ultrasound imaging. By combining a transfer learning–based CNN model with an image validation mechanism, the proposed framework provides a robust and interpretable solution for medical image analysis. The system demonstrates that integrating deep learning with validation and explainable AI significantly enhances detection accuracy and reliability compared to conventional standalone approaches. The MobileNetV2-based model effectively classifiesultrasoundimagesintonormalandkidney stone detected categories, capturing complex spatial patterns present in medical images. The inclusion of an image validation module ensures that only relevant kidney ultrasound images are processed, preventing incorrect predictions and improvingsystemrobustness.Furthermore,theuse ofGrad-CAM visualization enhances transparency byhighlightingregionsofinterestthatinfluencethe model’s decision, thereby increasing trust and interpretability in the diagnostic process. Evaluationusingstandardmetricssuchasaccuracy, precision, recall, F1-score, and confusion matrix analysis confirms the effectiveness and stability of theproposedsystem.Overall,KidneyAIachieves itsobjectivebydeliveringascalable,accurate,and user-friendly platform suitable for academic research and preliminary medical screening. Lookingahead,theKidneyAIsystemcanbefurther improved in several ways. Expanding the dataset with more diverse and high-quality ultrasound images can enhance model generalization and performance. Future work may explore advanced deep learning architectures such as EfficientNet, attention-based models, and ensemble learning techniques to further improve accuracy. The integrationofadditionalfeatures,suchasstonesize estimation, location detection, and severity analysis, can make the system more clinically useful. Incorporating advanced Explainable AI (XAI) techniques beyond Grad-CAM could provide deeper insights into model decisions. Additionally, deploying the system in real-time healthcare environments, optimizing performance using GPU acceleration, and integrating with hospital information systems can significantly enhance its practical applicability. The framework can also be extended to detect other kidney-related abnormalities, broadening its role in intelligent healthcare systems.

References

[1] Valarmathietal.,“Deeplearningframeworkfor kidneystoneidentificationusingCTimages,”Proc. Int. Conf. on Medical Imaging and Diagnostics, 2023. [2] Mangala Shetty et al., “CNN-based automated kidney stone detection using CT and ultrasound images,” Proc. Int. Conf. on Healthcare Informatics, 2024. [3] Kumar et al., “Multi-class kidney disease detection using deep learning models,” Proc. Int. Conf. on Artificial Intelligence in Medicine, 2024. [4] Al-Tbenawey et al., “Comparative study of KNN and CNN for kidney stone detection,” Proc. Int. Conf. on Image Processing and Applications, 2023. [5] George and Anita, “Kidney abnormality detection using CNN and SVM with ultrasound imaging,” Journal of Medical Imaging and Health Informatics, 2022. [6] Sabitha Rani et al., “Kidney stone detection usingwatershedsegmentationandprobabilistic neural networks,” Proc. Int. Conf. on Intelligent Computing Systems, 2023. [7] Tejbir Singh et al., “Ultrasound-based kidney stone localization using image segmentation techniques,”Proc. IEEE Int. Conf. on Advances in Medical Imaging, 2024. [8] Karthikeyanetal.,“DeepCNNarchitecturesfor kidney stone detection using CT datasets,” Proc. Int. Conf. on Deep Learning and Applications, 2024. [9] Shoayee Alotaibi, “Kernel-based machine learning approach for kidney stone identification using ultrasound images,” Journal of Biomedical Engineering, 2023. [10] ZahidaMir,“Real-timekidneystonedetection usingYOLO-basedobjectdetectionmodels,”Proc. IEEEInt.Conf.onComputerVisioninHealthcare, 2024. [11] Spoorthi B. Shetty, “Survey on kidney stone detection techniques using image processing and deeplearning,”Journal ofHealthcareEngineering, 2023. [12] Muhammed Altaf, “Comparative analysis of CNN and hybrid ML–DL models for kidney stone classification,”Proc.Int.Conf.onDataScienceand AI, 2024. [13] Noshi, “Web-based deep learning system for kidney stone detection using Flask deployment,” Proc. Int. Conf. on Smart Healthcare Systems, 2023. [14] Challapanati Vamsikrishna, “Ultrasound-basedkidneystonedetectionwithimagevalidation mechanisms,” Proc. Int. Conf. on Medical Image Analysis, 2024. [15] Eswar S., “Ensemble and transfer learning approaches for kidney stone classification,” Proc. Int. Conf.onMachineLearningApplications,2024.

Copyright

Copyright © 2026 Nandini ., Ambresh Bhadrashetty, Manjulabai Bhadrashetty. 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.