Table 7. Attention network method for SARS-CoV-2 analysis.

Author	Data type	Method	Data size	C	V	T and V size	Result	Key contribution and findings	Research gaps
Ullah et al. (2023)	X-rays image	Densely attention network	17,342	2	CV	T 80%: V 20%	AC: 97.22%, S: 96.87%, SP: 99.12%, P: 95.54%	Dense layers extract spatial features, channel attention builds weights, suppresses redundant representations.	Need to Expand model for pneumonia and other lung diseases to aid radio-logists.
Ouyang et al. (2020)	CTs image	Dual-sampling attention network	3,774	2	CV	T 80%: V 20%	AC: 87.5%, S: 86.9%, SP: 90.1%, F1: 82.0%.	Divide the areas affected by pneumonia infections to increase network focus and improve visual attention for more com- prehensible and interpretable models.	Prediction accuracy should be adaptive with large data
Yang et al. (2023)	X-ray image	Attention-based transformer	15,153	2	5K-CV	T 80%: V 20%	AC: 98.0%	Feature extraction and classification with Integrated Attention-based transformers model outperforms CNN in SARS-CoV-2 diagnosis.	Computational complexity
Wen et al. (2023)	X-ray image	Attentive capsule network	23,409	3	CV	T 80%: V 20%	AC: 96.3%, S: 98.8%, SP: 93.8%, ROC: 98.3%	Study proposes attention capsule sampling network for SARS-CoV-2 detection using key slices enhancement method on chest CT scans.	Inadequate feature extraction for individual slices or patient clinical information.
Christina Magneta, Sundar & Thanabal (2023)	X-ray image	Hierarchical attention network	NA	2	CV	T 80%: V 20%	AC: 93.36%	MRMVO-based HAN classifier incor- porates manta-ray foraging optimization and multi-verse optimizer, acquiring SARS-CoV-2 detection features from seg- mented lung lobes for targeted regions.	Time complexity need to be decreased
Hu et al. (2022)	X ray	Attention ResNet	4,449	3	CV	T 80%: V 20%	SN: 90.2%	Attention fusion feature for automatic SARS-CoV-2 from CT images	Result is biased to image segmentation
Fan et al. (2021)	X-ray image	Multi-kernel-size spatial channel attention	1,000	2	10K CV	T 80%: V 20%	AC: 98.2%	First stage extracts features, second stage uses multi-kernel attention modules, segmented pneumonia infection regions, for improved model interpretability and explainability.	Data modality and model generability should be improved
Zhang et al. (2021)	X-ray	Attention GAN	100	2	CV	T 80%: V 20%	S: 69.85% P: 94.6%	Hybrid model that performs good with some prepossessing	Worked on small data
Wang et al. (2021b)	CT scan	Attentive FGCN	320	2	CV	T 80%: V 20%	AC: 97.71%	Handles feature context efficiently	It is not adaptable enough