Table 8.

Approach-based comparative analysis of the proposed method with some selected related studies.

Studies	Approach	Domain of application	Differentiator with proposed study
78	Siamese convolutional neural network architecture called CNN-Siam, was applied to learn the feature representation of drug pairs from multimodal data of drugs	Prediction of drug-to-drug interactions (DDIs) based on modalities of chemical substructures, drug targets and enzymes	The RAdam and LookAhead optimization algorithms were relied on for improving accuracy based on ffeatures learned using the CNN-Siam, whereas our proposed model leverages a metaheuristic-based algorithm to select discriminant features learned using TwinCNN
41	Siamese neural network (SNN) is proposed for classification purpose in conjunction with k-nearest neighbour (k-NN) model	Applied to classification of MRI images samples for brain tumor detection	The study aimed at reducing feature space using shallow neural network as against the CNN architecture. On the contrary, our proposed TwinCNN is based on deep neural network capable of extracting rich features while a novel binary optimizer is applied for the feature space reduction
44	Deep learning-based Siamese neural network is design with attention mechanism	Detection of abnormality in product data at manufacturing site	The attention mechanism supports their feature extraction, it however introduces a very high representation of feature space Moreover, training of the model on small dataset contradicts our approach which leverages sufficient dataset to ensure that features space represents a good generalization
39	Triplet Siamese CNN based on benchmark architectures	Few-shot learning for detection of COVID-19	Ensemble of benchmark neural architectures were composed to build a triplet Siamese network. However, our proposed model is based on a dual neural architecture
45	Siamese CNN (SCNN) with minimal supervised learning	Applied for content-based retinopathy fundus image retrieval	Our proposed model combines features with predicted label for fusion which determines the multimodal classification
40	Siamese neural network based to enable one-shot classification	Handcrafted features were used to initiate the extraction of discriminant features	The study we propose leverages of binary optimizer with TwinCNN for feature extraction and selection of discriminant features
37	Siamese neural network for single modality image pair with two time points	Applied to monitor progression of disease	Our proposed TwinCNN is aimed at multimodal images combining histology and mammography samples
42	Twin CNN for extraction of feature maps based a content-based retrieval	Used for retrieval of Optical Coherent Tomography (OCT) scans	The TwinCNN proposed in our study is aimed for multimodality image classification with a novel feature extraction and reduction algorithm