Evolutionary Gravitational Neocognitron Neural Network-Based Blood Flow Velocity Prediction Using Multi-Exposure Laser Speckle Contrast Imaging

Wide-field and noncontact imaging technique called laser speckle contrast imaging (LSCI) is utilized to map blood flow. To greatly improve prediction accuracy, the block-matching along 3-dimensional transform domain collaborative filtering-dependent denoising technique is developed. To implement real-time denoising, the processing time makes complexity. Due to the existence of considerable noise and artifacts, this is challenging to achieve the acceptable level with less raw speckle images. Notwithstanding, it acts poorly while learning from the original LSCI speckle contrast images because of the uneven noise distribution. To overcome this issue, an evolutionary gravitational neocognitron neural network-based blood flow velocity prediction using multiple exposure laser speckle contrast imaging (EGNNN-BFVP-MeLSCI) is proposed. Initially, the input MeLSCI datasets are collected from real-time dataset. The input picture is enhanced and noise is removed using pre-processing technique called altered phase preserving dynamic range compression (APPDRC). Gray level co-occurrence matrix (GLCM) window adaptive approach-dependent feature extraction approach is used for the preprocessed pictures. Using GLCM window adaptive method, the picture characteristics, such as intensity information, images derivative, geodesic information, contrasts, energy, correlations, homogeneity, and entropy, are retrieved. Then, the extracted features are transferred into the EGNNN classifier for accurately predicting the blood flow velocity. The performance of proposed method is executed at python and evaluated under certain performance metrics, such as accuracy, precision, sensitivity, specificity, [Formula: see text]-measure, ROC, computation time, mean squared error, root mean squared error, predicted velocity. The proposed EGNNN-BFVP-MeLSCI method attains higher accuracy of 96.31%, 97.06%, and 93.52%, lower computational time of 97.22%, 91.39%, 96.41% compared with existing approaches, like DnCNN-BFVP-MeLSCI, CFD-BFVP-MeLSCI, DLS-BFVP-MeLSCI, CNN-BFVP-MeLSCI, DBN-BFVP-MeLSCI, CGAN-BFVP-MeLSCI and MVN-BFVP-MeLSCI, respectively.