lancet-header

Preprints with The Lancet is part of SSRN´s First Look, a place where journals identify content of interest prior to publication. Authors have opted in at submission to The Lancet family of journals to post their preprints on Preprints with The Lancet. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early stage research papers that have not been peer-reviewed. The findings should not be used for clinical or public health decision making and should not be presented to a lay audience without highlighting that they are preliminary and have not been peer-reviewed. For more information on this collaboration, see the comments published in The Lancet about the trial period, and our decision to make this a permanent offering, or visit The Lancet´s FAQ page, and for any feedback please contact preprints@lancet.com.

Comparison of the Diagnostic Performance of Deep Learning Algorithms for Reducing the Time Required for COVID-19 RT–PCR Testing

35 Pages Posted: 28 Mar 2022

See all articles by Yoonje Lee

Yoonje Lee

affiliation not provided to SSRN

Yu-Seop Kim

Hallym University

Da-in Lee

Hallym University

Seri Jeong

Hallym University - Kangnam Sacred Heart Hospital

Gu-Hyun Kang

Hallym University

Yong Soo Jang

Hallym University

Wonhee Kim

Hallym University

Hyun Young Choi

Hallym University

Jae Guk Kim

Hallym University

More...

Abstract

Background: The RAT has the advantage of a short test time and high diagnostic accuracy when performed on symptomatic people or on people with a high viral load, but its diagnostic accuracy is low for asymptomatic people. Its negative predictive value is also insufficient, which lowers the performance to detect true negatives. Because the dependability of negative results, i.e., the negative predictive value, is insufficient, there is a possibility that there will be an actual infected individual among those patients who had tested negative on the RAT. As several previous studies have proven, in the current situation, when the prevalence is rising, the RAT is not seen as an acceptable method to reduce the risk of infectious disease propagation. Every country is fully aware of such issues; however, the adoption of the RAT is unavoidable in the current situation because the spread of this infectious disease is constantly expanding.

Through a previous study, we recently developed and evaluated a deep learning algorithm that can reduce the COVID-19 RT–PCR diagnostic time. Therefore, based on our prior study, we constructed various deep learning algorithms to reduce the diagnostic time of RT–PCR by utilizing more data, and we compared the diagnostic performance of each algorithm and evaluated the clinical practical feasibility.

Methods: To train the models, the results of the RT–PCR virology tests were utilized as the reference. There were 1,270 positive RT–PCR results among the 2,540 patients whose data were included in the research, while 1,270 of the patients had negative results. These data were split into training and testing datasets. Curves of RT–PCR results from 1,000 positive and 1,000 negative cases were used to establish the data for the model training and validation. For testing, 270 positive and 270 negative cases were utilized. The raw data for the RT–PCR fluorescence values that were used in this study exhibited sequential characteristics that varied with the extraction time during the cycle. Thus, among the different deep learning models (“Recurrent Neural Network” (RNN), “Long Short-Term Memory” (LSTM), “Bidirectional Long Short-Term Memory” (Bi-LSTM), “Gated Recurrent Unit” (GRU), and “Transformer”) that were suitable for time series processing were used in this study.

Findings: The AUROC results appeared as follows in model No 10 according to each algorithm. The AUROC results were 85.2 for Bi-LSTM (95% CI, 82.2-88.1), 84.3 for GRU (95% CI, 81.6-87.1), 83.2 for LSTM (95% CI, 80.3-85.9), 79.4 for RNN (95% CI, 76.1-82.2), and 78.2 for transformer (95% CI, 74.9-81.7). In Model No 20, the AUROC values were 93.2 (95% CI, 91.0-95.0) for Bi-LSTM, 92.4 (95% CI, 90.3-94.5) for GRU, 92.0 (95% CI, 89.9-94.2) for LSTM, 91.9 (95% CI, 89.7-94.0) for RNN, and 91.9 (95% CI, 89.5-94.0) for transformer. In a pairwise comparison, there was no statistically significant difference in the AUROC values of the algorithms in Models No 10 and 20 (all p > 0.05).

Interpretation: Among the five deep learning algorithms capable of training time series data, Bi-LSTM and GRU were shown to be able to decrease the time required for RT–PCR diagnosis by half or by 25% without significantly impairing the diagnostic performance of the COVID-19 RT–PCR test.

Funding Information: This research was supported by a grant from Hallym University Research Fund 2020 (HURF-2020-64) and National Research Foundation of Korea Grant Fund, the Korean Government: NRF-2019R1C1C1007890

Declaration of Interests: Seri Jeong, the corresponding author of this study, declares that there are no conflicts of interest related to this paper. The other authors declare that there are no conflicts of interest associated with this study.

Ethics Approval Statement: This study was approved by the Institutional Review Committee (HKS 2020-07-007) of Hallym University Kangnam Sacred Heart Hospital in Korea; the requirement for informed consent was waived because the subjects' data were anonymized. This study was conducted in accordance with the STARD guidelines and regulations for a study related to the diagnostic accuracy of COVID-19 RT–PCR.

Keywords: COVID-19, SARS-CoV2, RT-PCR, Deep learning, Omicron, Pandemic

Suggested Citation

Lee, Yoonje and Kim, Yu-Seop and Lee, Da-in and Jeong, Seri and Kang, Gu-Hyun and Jang, Yong Soo and Kim, Wonhee and Choi, Hyun Young and Kim, Jae Guk, Comparison of the Diagnostic Performance of Deep Learning Algorithms for Reducing the Time Required for COVID-19 RT–PCR Testing. Available at SSRN: https://ssrn.com/abstract=4068372 or http://dx.doi.org/10.2139/ssrn.4068372

Yoonje Lee

affiliation not provided to SSRN ( email )

No Address Available

Yu-Seop Kim

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Da-in Lee

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Seri Jeong (Contact Author)

Hallym University - Kangnam Sacred Heart Hospital ( email )

Seoul
Korea, Republic of (South Korea)

Gu-Hyun Kang

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Yong Soo Jang

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Wonhee Kim

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Hyun Young Choi

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Jae Guk Kim

Hallym University ( email )

39 Hallymdaehak-gil
Chuncheon, Gangwon-do, 200-702
Korea, Republic of (South Korea)

Click here to go to TheLancet.com

Paper statistics

Downloads
32
Abstract Views
229
PlumX Metrics