#0197

Deep Neural Network and Machine Learning Radiomic Model for Renal Tumors as Accuracy in Diagnosis and Operative Planning

E. Negara¹, B. Daryanto¹, K. Seputra¹, T. Budaya¹, I. Irmawati², M. Pratiwi²

¹Universitas Brawijaya, Department of Urology, Malang, Indonesia
²Indonesia International Institute for Life Science, Faculty of Bioinformatic, Jakarta, Indonesia

Introduction:

Kidney disease is a critical health issue that demands accurate diagnosis and prompt medical treatment. As we move toward the 6.0 industrial revolution, the fusion of human intelligence, artificial intelligence (AI), and big data. This has created a demand for technologies that can streamline and accelerate image interpretation and surgical decision-making. Artificial intelligence, especially Convolutional Neural Network, shows significant promise in tackling this issue. CNNs, a machine learning approach, have proven effective in image analysis and can be trained to identify intricate patterns in medical images with remarkable accuracy. The integration of CNN into medical imaging systems aims to create a model that can not only quickly and accurately identify tumor markers but also expedite the diagnosis and surgical decision-making process.

Material and methods:

This study will utilize a dataset comprising CT-Scan and MRI scans annotated by radiologists, along with data from PACS Bangladesh and Saiful Anwar Hospital. The CNN model to be developed seeks to identify and distinguish different types of renal tumors. This research consists of several stages: data pre-processing, model training with the optimal CNN architecture, and model validation using accuracy, sensitivity, and specificity metrics. It is hoped that the developed model will not only be able to detect kidney disease with high accuracy but also provide new insights into medical image interpretation, with great potential for integration into clinical diagnostic systems.

Results:

The experimental results indicate that the CT-based kidney disease detection model achieves a validation accuracy of 99.97%. In comparison, the ViT model achieved 97.44% accuracy on MRI image data, while the DeiT model reached 99.43%, and the Swin model attained 99.72%. These models not only show strong performance in identifying kidney tumors but also offer valuable insights into the interpretation of medical images. With the potential for integration into clinical diagnostic systems, this research significantly contributes to the progress of medical diagnostic technologies, particularly in enhancing the effectiveness and efficiency of kidney disease management.