Artificial intelligence with deep convolutional neural network-based clinical decision-making on kidney Oncology using multimodal imaging.

Kidney cancer is among the most prevalent malignancies globally. Accurate analysis is vital in managing kidney cancer patients affected by tumour volume or size, cancer stages, and types. This results in a massive quantity of medical data on every patient, like histopathological images, magnetic resonance imaging (MRI) or computed tomography (CT) scans, and other clinical information. In recent times, automatic analysis of kidney cancer has been a significant task, particularly when utilizing deep learning (DL), owing to the import of training medicinal data are demanding and costly to acquire. Unlike preceding generations of machine learning (ML), DL approaches based on convolutional neural networks (CNNs) can deal with raw intensity images and acquire to extract analytical features automatically. This study uses multimodal imaging to propose an Artificial Intelligence with Deep Convolutional Neural Network Based Clinical Decision Making on Kidney Oncology (AIDCNN-CDMKO) methodology. The primary purpose of the AIDCNN-CDMKO methodology framework is to support clinical decision-making for the precise identification and classification of kidney tumours. The presented AIDCNN-CDMKO methodology initially utilizes the Gaussian filter (GF) in the image pre-processing to enhance image clarity by reducing noise and improving feature detectability. Furthermore, the fusion of feature extraction models such as VGG16, MobileNetV1, and EfficientNetB5 is employed to capture rich and complementary representations from multimodal imaging data. The attention mechanism-based CNN and bi-directional long short-term memory (CNN-BiLSTM-AM) classifier is utilized for kidney cancer detection. To validate the improved analytical results of the AIDCNN-CDMKO approach, extensive simulations are conducted under the CT Kidney and RCC Kidney Histopathology datasets. The comparative study of the AIDCNN-CDMKO exemplified a superior accuracy value of 98.33% and 98.42% under the dual datasets.