Table 2 Inferences from literature Review.

Sequential CNN

Fuzzy CNN⁵, Confluence CNN¹², ESDPCOS¹³, AMCNN¹⁴, KNN based CNN¹⁵, MLOD¹⁶, Ocys-Net¹⁷, ITL-CNN¹⁹, Ensemble CNN²⁰, Hybrid CNN²³, Sequential CNN²⁷, 2D CNN²⁹, Tri-stage wrapper CNN³¹, SSFSE-DL³⁶, DLNNSVM⁴⁰, GrabCut and Fuzzy Logic -SNN Model⁴¹

Best in handling images with permissible noise and scalable to apply the same model to any large number of datasets

Easy to integrate with other CNN architectures or combined with TL

Image pixels and the ROI are identified uniquely in each convolution layer operation

The model should be refined with labelled data

Need to validate with various cross-validation methods to fine-tune the accuracy

Need separate validation to handle augmentation images

Depends on feature filtering to extract deep features from the ultrasound images

Recurrent CNN

Complex Spatial Recurrent Neural Network U-Net²², Elman NN³², Back Propagation Algorithm³⁵

Spatial relationship between image pixels is effectively correlated

Relationship between the image frames is reused

Effectively extract the patterns in ultrasound image

Best suited for large dataset

Less performance while handling image with ling sequences and cannot handle static image data

Depends on effective feature extraction methods

Pre-Trained CNN

AResUNet¹, Inception CNN², ResNet⁴, VGGNet⁴, Inception V3^4,28, PCOS-WaveConvNet⁶, PCONet⁷, VGG16^8,9,24,30, ASPPNet¹⁰, HHO-DQN¹⁸, SqueezeNet²⁶, UNet and EfficientNet³⁸, GIST-MDR⁴², U-Net³ and ResNet³

AResUNet achieves 98% of accuracy

Inception CNN achieves 84.81% of accuracy

Exhibits good performance with little amount of data

Best in extracting the deep features

Model need to be fine-tuned with several activation function and number of layers to improve the performance

Depends on balanced data before fitting the model