BIO3011 Research Methods in Biology
Task:
Introduction
Artificial Intelligence (AI) strategies overall and Convolution Neural Networks (CNN) specifically have achieved victories in clinical picture examination and characterization. A few groups have partaken in tackling the issue of foreseeing Covid-19 through different techniques. To recognize COVID-19 patients from their chest X-Ray images, Abbas et al. [1] presented and authorized a sophisticated convolution neural network named DeTraC- Decompose, Transfer, and Compose. He presented a decomposition tool to check for abnormalities in the dataset by examining class boundaries in order to get a high amount of exactness and affectability For diagnosing Covid-19, Wang et al. developed an exchange learning technique (Xception model) based on deep learning models. The proposed technique has a diagnosis accuracy of 96.75 %. Deep components and AI arrangements were also used to develop a competent analytic method for improving the Xception model's precision by. The designers claimed that their offered technique resulted in greater arrangement accuracy and effective symptomatic execution of Covid-19 based on the findings. In any event, the creators have not made any comparisons between their results with existing comparable efforts.
Three profound exchange models, AlexNet, GoogleNet, and ResNet, were used on a dataset of 307 images with four different types of classes: Covid-19, typical pneumonia bacterial, and pneumonia infection, in the suggested model by Loey et al. [5]. To reduce memory use and execution time, the investigation activity was divided into three scenarios. Last but not least, there was a deep trade model. GoogleNet was able to achieve testing precision and approval exactness. Another team from the University of Saskatchewan, Canada used DL-based CNN and compared their performance. This comparison had done with ResNet, Xception, inception models. This trail had done with multiple CNN models. Therefore Xception model gives upmost perfection of, but the disadvantage was it can be done for the lesser dataset.
A team from the Zhejiang Institute of Mechanical and Electrical Engineering, casserole, used CT scans to apply transfer learning to Covid-19 testing. This model, which was pre-trained on ImageNet21k, has generalizability in CT images and a fineness of for recognising Covid-19 instances.
Drawbacks
CT scan pictures have a high sensitivity in identifying and detecting instances of COVID-19 when compared to RT-PCR; however, their specificity is limited, hence it is advised that if medical imaging is required.
When compared to the models stated above, such as AlexNet, Xception, and InceptionV3, Overall performance is significantly better with Xception, although the sample is much less.
