Hybrid Convolutional Neural Network and Extreme Learning Machine for Kidney Stone Detection

• Published in: 2023 Second International Conference on Electronics and Renewable Systems (ICEARS) pp. 936 - 942
• Main Authors: G, Pandiya Rajan, Sapra, Puneet, Mary, S. Suma Christal, Chauhan, Amit, Parte, Smita Athanere, Nishant, Neerav
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02.03.2023
• Subjects: Convolutional Neural Network (CNN); Extreme Learning Machine (ELM); Feature extraction; Histograms; Image segmentation; Speckle; Support vector machines; Training; Ultrasonic imaging
• DOI: 10.1109/ICEARS56392.2023.10085243

When it comes to diagnosing structural abnormalities including cysts, stones, cancer, congenital malformations, swelling, blocking of urine flow, etc., ultrasound imaging plays a key role in the medical sector. Kidney detection is tough due to the presence of speckle noise and low contrast in ultrasound pictures. This study presents the design and implementation of a system for extracting kidney structures from ultrasound pictures for use in medical procedures such as punctures. To begin, a restored input image is used as a starting point. After that, a Gabor filter is used to lessen the impact of the speckle noise and refine the final image. Improving image quality with histogram equalization. Cell segmentation and area based segmentation were chosen as the two segmentation methods to compare in this investigation. When extracting renal regions, the region-based segmentation is applied to obtain optimal results. Finally, this study refines the segmentation and clip off just the kidney area and training the model by using CNN-ELM model. This method produces an accuracy of about 98.5%, which outperforms CNN and ELM models.