Targeted Self Supervision For Classification On A Small Covid-19 Ct Scan Dataset

Abstract

Traditionally, convolutional neural networks need large amounts of data labelled by humans to train. Self supervision has been proposed as a method of dealing with small amounts of labelled data. The aim of this study is to determine whether self supervision can increase classification performance on a small COVID-19 CT scan dataset. This study also aims to determine whether the proposed self supervision strategy, targeted self supervision, is a viable option for a COVID-19 imaging dataset. A total of 10 experiments are run comparing the classification performance of the proposed method of self supervision with different amounts of data. The experiments run with the proposed self supervision strategy perform significantly better than their non-self supervised counterparts. We get almost 8% increase in accuracy with full self supervision when compared to no self supervision. The results suggest that self supervision can improve classification performance on a small COVID-19 CT scan dataset. Code for targeted self supervision can be found at this link: https://github.com/Mewtwo/Targeted-Self-Supervision/tree/main/COVID-CT

A self-supervised COVID-19 CT recognition system with multiple regularizations

“…While the unsupervised pretraining implemented in CSSL can alleviate this problem, due to the fact that it only uses data instances to mine useful information. And CSSL has achieved good recognition results for the COVID-CT dataset [ 8 , 16 , 17 ]. In general, CSSL-based pretraining is less prone to overfitting.…”

“…Huang et al point out that the method based on deep learning is difficult to deal with imprecise and uncertain information due to the low contrast of CT images, so they develop a classification network based on belief function using semi-supervised learning [ 9 ]. Ewen et al propose a targeted self-supervised method, which makes the network architecture used by pretext tasks for self-supervision and downstream tasks unchanged, simplifying the experimental process, and enabling all layers of the network to gain benefits from self-supervised learning [ 17 ]. Mishra et al combine the prediction results of several different deep CNN models to identify COVID-19 [ 13 ].…”

“…Therefore, utilizing CT images to diagnose COVID-19 is a feasible and promising solution. Some studies have developed related COVID-19 diagnostic methods on basis of CT images [ [7] , [8] , [9] , [10] , [11] , [12] , [13] , [14] , [15] , [16] , [17] ]. The research work conducted in Refs.…”

Quantifying prognosis severity of COVID-19 patients from deep learning based analysis of CT chest images

“…Also, multiple studies have used COVID-CT as their dataset for diagnosis analysis because of its high patient count of CT chest scan slice images. And these studies have used transfer learning, capsule networks, AutoML, self-supervision and novel cross validation approaches to achieve better results in comparison to baseline solutions [12,16,26,28,33]. But, most of the above stated studies haven't employed extensive techniques to remove data related biases in predictions and also haven't analyzed limitations of the data quality.…”

A k-mer Based Approach for SARS-CoV-2 Variant Identification

“…Massive testing and largescale sequencing produced a huge amount of data, creating ample opportunity for the bioinformatics community. Researchers started exploring the evolution of SARS-CoV-2 [16] to vaccine landscapes [34] and long-term effects of covid to patients [35]. In [25], the authors indicate how the coronavirus spike protein is fine-tuned towards the temperature and protease conditions of the airways, to enhance virus transmission and pathology.…”