Design and validation of renal stone detection using multi-architecture feature extraction with deep sequential learning model on axial computed tomography images.

Kidney stone disease is a significant public health threat, with its prevalence escalating due to evolving dietary habits, rising rates of obesity, other medical conditions, and the use of certain supplements. A kidney stone, otherwise known as a renal calculus, is a solid mass of crystallized minerals that aggregates within the kidneys. The proper identification of this renal condition is vital because it represents a serious health issue that requires accurate detection for effective treatment. Imaging techniques play a vital role in diagnosing kidney diseases, including kidney stones. Computed tomography (CT) is among the imaging techniques utilized to detect kidney stones by medical specialists. CT scans provide information on a stone's specific location and size, allowing for an estimation of the chances for natural expulsion, thus potentially avoiding the need for surgical procedures. Deep learning (DL) models are progressively renowned as a robust tool for disease diagnosis in the biomedical domain. This study presents a Feature Integration and Sequential Attention Framework for Kidney Stone Detection (FISAF-KSD) approach. The primary goal of this work is to develop a reliable and efficient system that can accurately identify kidney stones from CT images. To achieve this, the FISAF-KSD approach initially performs image pre-processing and augmentation to improve input image quality and prepare CT images for further analysis. Following this, feature extraction is carried out through a fusion of three DL models, such as EfficientNetV2L, InceptionV3, and ResNet-101, to capture the key features of kidney stones at both detailed and broad levels. Finally, a bidirectional gated recurrent unit network (BiGRU) with an attention mechanism (AM) is employed to classify renal stones effectively. The performance analysis of the FISAF-KSD methodology is thoroughly examined under the Axial CT imaging dataset. The FISAF-KSD methodology accomplished [Formula: see text] of 98.75%, [Formula: see text] of 98.76%, [Formula: see text] of 98.75%, [Formula: see text] of 98.75%, and [Formula: see text] of 98.75%. The results indicate that the FISAF-KSD methodology performed better compared to existing approaches.