A considerable amount of studies has been done to develop biometric systems for brain stroke and face recognition using different techniques. This section provides a review of studies that adopted traditional ML and DL approaches in the biometric systems.

Review on Brain Stroke

Maier et al. [20] applied nine classification methods, including generalized linear models, random decision forests (RDFs), and CNNs, to order ischaemic stroke and inferred that RDFs and CNNs can give preferred grouping exactness over different strategies. Another study [21] presented a forecast model with DT, ANN, SVM, logistic regression (LR), and ensemble approach generalized boosted model (GBM) to foresee ICU move of stroke patients and inferred that GBM gave the most elevated precision. Kansadub et al. [22] used DTs, naive Bayes, CNN, and ANN to anticipate stroke and reasoned that DT yielded preferable order over different techniques. Sung et al. [23] used kNN, multiple linear regression, and a regression tree model to predict the stroke severity index and exhibited that к-nearest neighbour (kNN) has preferred precision over different models.

Kumar et al. [24] studied the performance of the implemented approach offered higher accuracy for the three-class classification problem which also solved the complexity in segmentation problems. A robust technique for the automatic segmentation of haemorrhage, ischaemic stroke, and tumour lesions from the MRI and CT brain images was contrived by using the Decision Tree characterization model. Snehkunj et al. [25] focused on the feature extraction of the MRI and CT brain images. The abnormalities such as brain haemorrhage and brain tumour were considered into account, which were diagnosed using the same methodology. Various phases were explored such as brain image extraction, transformation, and progression of the MRI or CT images. The accuracy of detecting the abnormalities in the images was enhanced. Ferdian [26] used a robust and accurate segmentation method based on a combination of an atlas-based and active contours segmentation. The experimental analysis revealed an extraordinary correlation with increased accuracy and was well suited for the reliable ventricle segmentation in stroke patients.

Few scientists are working on stroke expectation with ML calculations. Massive research commitments are depicted in this segment. A past report utilised ANN strategy, prepared with six diverse multilayer perceptron calculations to anticipate the mortality of stroke patients which created a precision of 80.7% [27]. Another study utilised SVM, kNN, and ANN to mechanise the discovery of ischaemic stroke, which recommended that SVM has higher expectation precision [28,29]. Amini et al. [30] anticipated stroke rate by utilising к-nearest neighbour and 4.5 decision tree techniques to uncover that C4.5 decision tree strategies yielded a higher exactness rate of 95.42%. Another group [31] used ML methods and SVM to predict stroke thrombolysis result, which indicated that SVM was more accurate. Cheng et al. [32] predicted ischaemic stroke utilising two ANN models that gave an accuracy rate of 79.2% and 95.1%.

Priya et al.’s [33] study predicts the sort of stroke for a patient dependent on classification methodologies. The classes of SVM and ensemble (bagged) gave 91% accuracy with 0.0000 negative predictive value, while ANN prepared with the stochastic gradient descent algorithm outperformed other algorithms, with a higher classification accuracy of 95% with a lower standard deviation of 14.69. Chantamit-O-Pas et al. [34] propose a stroke forecast through DL. The information on clinical area issues couldn’t be followed precisely by the conventional prescient models.

Li et al. used generalised linear model, Bayes model, and decision tree model to predict the risk of ischaemic stroke and other thromboembolism of individual with atrial fibrillation [35]. Zhang et al. utilised an assortment of filter-based component choice models to improve the incapable element determination in the existing exploration on stroke hazard recognition [36].

Accurate classification and sensible intercession for high-chance populace can successfully lessen the weight of stroke on families and the society. It is important to consider the review of the grouping model to guarantee the congruity of stroke mediation [37,38]. Al-Maqaleh et al. used decision tree, Naive Bayesian, and neural system to predict the coronary illness and compared their exhibition in terms of accuracy [39]. Table 14.2 shows the summary of different ML techniques used for brain stroke detection and classification.

< Prev   CONTENTS   Source   Next >