A novel approach for automated diagnosis of kidney stones from CT images using optimized InceptionV4 based on combined dwarf mongoose optimizer

In this study, a novel methodology is proposed to combine deep learning techniques and metaheuristic optimization to enable early diagnosis of patients with kidney stone using CT images. This approach focuses on the utilization of the InceptionV4 model, a state-of-the-art deep learning architecture known for its outstanding performance in image classification tasks. To further enhance the performance of InceptionV4, a modified metaheuristic algorithm called Combined Dwarf Mongoose Optimization (CDMO) algorithm is introduced. The CDMO algorithm optimizes the arrangement and selection of hyperparameters for the InceptionV4 model. By fine-tuning the model’s arrangement, the aim is to improve its accuracy and efficiency in detecting kidney stones from CT images. This optimization process ensures that the InceptionV4 model is tailored specifically to the task of kidney stone diagnosis, maximizing its diagnostic capabilities. To evaluate the performance of the proposed methodology, it is applied to the CT Kidney Dataset, which contains a diverse range of CT images with and without kidney stones. The accuracy of the methodology is compared with existing methodologies, including Kronecker product-based Convolution Neural Network (K-CNN), Combined Discrete First Chebyshev Wavelets Transform and Convolutional Neural Network (DFCWTCNN), Enhanced Deep Neural Network (EDNN), hybrid Convolutional Neural Network and Extreme Learning Machine (CNN-ELM), and Probabilistic Neural Network (PNN) and Watershed Algorithm (PNN/WA). The results demonstrate the superior performance of the proposed method compared with other methods in terms of accuracy, F1-score, and Jaccard Index. These results provide evidence of the system's dependability, efficacy, and potential for use in medical picture classification tasks. The classification of CT scans as positive or negative for kidney stones demonstrates a notable level of accuracy, reaching 98.14%. This high accuracy is attributed to the model's sensitivity of 97.74% and specificity of 96.33%. Additionally, the model has a precision rate of 86.66%, demonstrating its ability to accurately forecast positive instances. Moreover, it maintains a balanced trade-off between sensitivity and accuracy, as shown by its F1-score of 92.51%. By enabling early diagnosis of kidney stones, the proposed methodology has the potential to improve patient's outcomes and reduce healthcare costs associated with delayed diagnosis and complications.