Advancements in Kidney Disease Diagnosis: The Role of Deep Learning Techniques in Modern Medicine
Pratham Kaushik, Savinder Kaur
Abstract
The detection of kidney diseases is essential for early clinical intervention and effective therapy. In this paper, a classification model is described for detecting four types of kidney conditions: normal, cyst, tumor, and stone. This model was applied by training a convolutional neural network on a diverse dataset to detect a high degree of accuracy and reliability. The performance metrics of the model for classification are evaluated in terms of precision, recall, and F1-score. For class Cyst, the model obtained a precision of 89.64%, a recall of 88.56%, and an F1-score of 89.10% with the support of 556 instances. The Normal class obtained metrics with an accuracy of 90.00%, a recall of 89.87%, and an F1-score of $\mathbf{8 9. 9 3 \%}$ across the support of 762 cases. Lastly, on the Stone class, the model added a precision of 86.26%, a recall of 89.52%, and an F1-score of 87.86% from the 207 instances. Tumor class: $\mathbf{9 0. 0 0 \%}$ precision, recall, and $F 1$-score for 342 instances. It achieved an accuracy of 90.46% on 1867 instances in total. Its macro average precision, recall, and F1-score were 88.97%, 89.49%, and 89.22%, respectively, while the weighted average precision, recall, and F1-score were $89.48 \%, 89.46 \%$, and $\mathbf{8 9. 4 7 \%}$. Most importantly, these results demonstrate that this proposed classification approach is very practical for diagnosis concerning kidney diseases and thus greatly aids the medical fraternity in clinical decision-making.