Adapting Convolutional Neural Networks for Chronic Kidney Disease Detection from Structured Data

Abstract

A progressive deterioration in kidney function that often goes undiagnosed until it reaches severe stages is the hallmark of chronic kidney disease (CKD), a worldwide health concern. Preventing the development of end-stage renal disease requires an early and precise diagnosis. Conventional diagnosis techniques mostly depend on clinical knowledge and laboratory analysis, which may be laborious and unreliable in resource-constrained environments. In this work, we use the structured, tabular data from the UCI CKD dataset to suggest a novel use of Convolutional Neural Networks (CNNs) for CKD classification. By transforming the input into two-dimensional grids, CNNs may now be used to simulate spatial connections in structured datasets, despite their usual employment in image-based tasks. In order to capture feature interactions and enable the CNN to learn intricate patterns linked to CKD diagnosis, our method preprocessed the clinical data and organised them into a 2D matrix format. With an average accuracy of 96.1%, our CNN-based model proved that convolutional architectures are capable of efficiently classifying chronic kidney disease (CKD) from structured clinical data. These findings imply that CNNs, even in the absence of picture data, may be a potent substitute for conventional machine learning models in healthcare applications. The adaptability of deep learning methods and their potential to facilitate prompt, automated CKD diagnosis in clinical decision-making systems are shown by this work.

Introduction

The study aims to develop a Convolutional Neural Network (CNN) model to automatically detect Chronic Kidney Disease (CKD) using structured clinical data. It leverages the CNN's ability to learn complex feature patterns and aims to demonstrate its effectiveness even outside traditional image-based domains.

Background

• CKD is a silent, progressive disease affecting millions worldwide, often detected late.

• Early detection is critical to prevent complications like cardiovascular disease, anemia, or kidney failure.

• Traditional diagnosis methods are effective but time-consuming, error-prone, and not scalable.

• Machine learning and particularly deep learning (CNNs) are explored for automating CKD classification using structured data.

Dataset

• UCI CKD Dataset: 400 patient records, each with 24 clinical features (11 numerical, 13 categorical).

• Target: Binary classification (CKD or not).

• Features include blood pressure, albumin, hemoglobin, creatinine, etc.

Methodology

1. Preprocessing

• Missing values imputed (mode for categorical, mean/median for numeric).

• Label encoding for categorical features.

• Min-Max normalization applied.

• Data reshaped into 3×3 matrices (padded with zeros) to make it suitable for CNN input.

2. CNN Architecture

• Input: 3×3×1 grid of features.

• Layers:

  • Convolutional Layer (32 filters, ReLU)

  • Max Pooling

  • Dense Layer (64 neurons, Dropout 0.3)

  • Output Layer (1 neuron, Sigmoid for binary classification)

• Optimizer: Adam

• Loss: Binary Cross-Entropy

• Training: 60 epochs, early stopping based on validation loss.

3. Implementation & Deployment

• Developed using Python (TensorFlow, Keras, Scikit-learn).

• Deployed via a Flask web interface, allowing users to input clinical data and receive CKD predictions with confidence scores.

• Interface suitable for healthcare workers or patient self-assessment.

Performance & Results

• Accuracy: 96.1%

• Precision: 95.6%

• Recall: 96.8%

• F1-Score: 96.2%

• AUC-ROC: 0.97

• 10-fold cross-validation confirmed model robustness and generalization.

• Confusion Matrix: Few misclassifications (1 false negative, 2 false positives).

Comparative Analysis

• CNN vs Traditional Models:

  • CNN outperformed Decision Trees, SVMs, and Logistic Regression.

  • Traditional models required manual feature engineering; CNN learned feature importance automatically.

• Feature Importance:

  • The CNN gave higher weight to key clinical features like serum creatinine, albumin, and hemoglobin, aligning with medical domain knowledge.

Conclusion

Based only on structured clinical data from the UCI CKD dataset, this research showed how well a Convolutional Neural Network (CNN) can diagnose Chronic Kidney Disease (CKD). The CNN was able to discover spatial correlations between clinical characteristics that conventional models could miss by converting tabular patient data into a two-dimensional grid format. With an average accuracy of 96.1% in testing, precision of 95.6%, recall of 96.8%, and an AUC-ROC of 0.97 over many validation folds, the model demonstrated strong performance. These findings demonstrate CNNs\' promise in structured dataset medical classification challenges as well as image-based applications. A reliable and scalable method for diagnosing CKD was made possible by CNNs\' automated feature extraction and local pattern learning capabilities. This method lessens the need for intensive feature engineering and might be the basis for upcoming clinical decision support technologies that operate in real time. The suggested CNN-based CKD diagnosis method has substantial practical usefulness in patient-centered treatment in addition to its technological advantages. In order for patients to engage with the system on their own and have a better understanding of their health state, nurses may play a crucial role in teaching them how to utilise the user interface. Making the tool user-friendly and accessible encourages patients to take an active role in keeping an eye on their kidney health. By empowering people to seek medical care sooner, early awareness raised by this application may help initiate therapy at an earlier stage of chronic kidney disease (CKD) and improve long-term results. This combination of AI and patient education promotes a healthcare strategy that is more interactive and preventative. Although the CNN-based methodology produced encouraging outcomes, there remains room for development and expansion. To improve model generalisation, future research might concentrate on expanding the dataset to encompass a bigger and more varied patient group. Investigating other grid configurations or encoding methods for tabular data might potentially enhance the model\'s ability to understand feature relationships. Furthermore, using interpretability strategies like gradient-based attribution or saliency maps might provide more profound understanding of the clinical characteristics that the CNN most often uses, boosting openness and confidence in healthcare environments. A useful first step towards clinical adoption would be incorporating this model with real-time data input into a hospital system or cloud-based application.

References

[1] Francis, A., Harhay, M.N., Ong, A.C.M. et al. Chronic kidney disease and the global public health agenda: an international consensus. Nat Rev Nephrol 20, 473–485 (2024). https://doi.org/10.1038/s41581-024-00820-6. [2] Bajwa J, Munir U, Nori A, Williams B. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J. 2021 Jul;8(2):e188-e194. doi: 10.7861/fhj.2021-0095. PMID: 34286183; PMCID: PMC8285156. [3] Patil, Dimple & Rane, Nitin & Desai, Pravin & Rane, Jayesh. (2024). Machine learning and deep learning: Methods, techniques, applications, challenges, and future research opportunities. 10.70593/978-81-981367-4-9_2. [4] Sugiyarti, E.; Jasmi, K.A.; Basiron, B.; Huda, M.; Shankar, K.; Maseleno, A. Decision Support System of Scholarship Grantee Selection using Data Mining. Int. J. Pure Appl. Math. 2018, 119, 2239–2249. [5] Aljaaf, A.J.; Al-Jumeily, D.; Haglan, H.M.; Alloghani, M.; Baker, T.; Hussain, A.J.; Mustafina, J. Early Prediction of Chronic Kidney Disease Using Machine Learning Supported by Predictive Analytics. In Proceedings of the 2018 IEEE Congress on Evolutionary Computation (CEC), Rio de Janeiro, Brazil, 8–13 July 2018; pp. 1–9. [6] Gorzaáczany, M.B.; Rudzi?ski, F. Interpretable and accurate medical data classification—A multi-objective genetic-fuzzy optimization approach. Expert Syst. Appl. 2017, 71, 26–39. [7] Kunwar, V.; Chandel, K.; Sabitha, A.S.; Bansal, A. Chronic Kidney Disease analysis using data mining classification techniques. In Proceedings of the 2016 6th International Conference—Cloud System and Big Data Engineering (Confluence), Noida, India, 14–15 January 2016; pp. 300–305. [8] Md. Ariful Islam, Md. Ziaul Hasan Majumder, Md. Alomgeer Hussein, Chronic kidney disease prediction based on machine learning algorithms, Journal of Pathology Informatics, Volume 14, 2023, 100189, ISSN 2153-3539, https://doi.org/10.1016/j.jpi.2023.100189. [9] Rahul Sawhney, Aabha Malik, Shilpi Sharma, Vipul Narayan, A comparative assessment of artificial intelligence models used for early prediction and evaluation of chronic kidney disease, Decision Analytics Journal, Volume 6, 2023, 100169, ISSN 2772-6622, https://doi.org/10.1016/j.dajour.2023.100169. [10] Alsekait, D.M.; Saleh, H.; Gabralla, L.A.; Alnowaiser, K.; El-Sappagh, S.; Sahal, R.; El-Rashidy, N. Toward Comprehensive Chronic Kidney Disease Prediction Based on Ensemble Deep Learning Models. Appl. Sci. 2023, 13, 3937. https://doi.org/10.3390/app13063937. [11] Arif, M.S.; Mukheimer, A.; Asif, D. Enhancing the Early Detection of Chronic Kidney Disease: A Robust Machine Learning Model. Big Data Cogn. Comput. 2023, 7, 144. https://doi.org/10.3390/bdcc7030144. [12] Rubini, L., Soundarapandian, P., & Eswaran, P. (2015). Chronic Kidney Disease [Dataset]. UCI Machine Learning Repository. https://doi.org/10.24432/C5G020. [13] Singla R, Ringstrom C, Hu G, Lessoway V, Reid J, Nguan C, Rohling R. The open kidney ultrasound data set. arXiv preprint arXiv:2206.06657. 2022 Jun 14. [14] Afkanpour M, Tehrany Dehkordy D, Momeni M, Tabesh H. Conceptual framework as a guide to choose appropriate imputation method for missing values in a clinical structured dataset. BMC Med Res Methodol. 2025 Feb 20;25(1):43. doi: 10.1186/s12874-025-02496-3. PMID: 39979819; PMCID: PMC11843774. [15] Dina Elreedy, Amir F. Atiya, A Comprehensive Analysis of Synthetic Minority Oversampling Technique (SMOTE) for handling class imbalance, Information Sciences, Volume 505, 2019, Pages 32-64, ISSN 0020-0255, https://doi.org/10.1016/j.ins.2019.07.070.

Copyright

Copyright © 2025 Neethiselvam Devadoss. 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.