loading page

CAN DEEP LEARNING PREDICT LABORATORY VALUES IN COVID-19?
  • +8
  • NAZLIM AKTUĞ DEMİR,
  • Onur Ural,
  • Asli Ural,
  • Sua Sumer,
  • Hatice Esranur Kiratli,
  • Lutfi Saltuk Demir,
  • Ediz Uslu,
  • Fikret Kanat,
  • Ugur Arslan,
  • Husamettin Vatansev,
  • Hakan Cebeci
NAZLIM AKTUĞ DEMİR
Selcuk University Faculty of Medicine
Author Profile
Onur Ural
Selcuk University Faculty of Medicine
Author Profile
Asli Ural
JotForm Yazilim A.S.
Author Profile
Sua Sumer
Selcuk University Faculty of Medicine
Author Profile
Hatice Esranur Kiratli
Selcuk University Faculty of Medicine
Author Profile
Lutfi Saltuk Demir
Necmettin Erbakan University Faculty of Medicine
Author Profile
Ediz Uslu
JotForm Yazilim A.S.
Author Profile
Fikret Kanat
Selcuk University, Faculty of Medicine
Author Profile
Ugur Arslan
Selcuk University Faculty of Medicine
Author Profile
Husamettin Vatansev
Selcuk University Faculty of Medicine
Author Profile
Hakan Cebeci
Selcuk University Faculty of Medicine
Author Profile

Abstract

Aims: Laboratory findings in COVID-19 patients vary according to the severity of the disease. This study aimed at defining a system of formulas that may predict the presence of thoracic CT involvement, the extent of such involvement and the need for intensive care stay on the basis of patient laboratory data using the Waikato Environment for Knowledge Analysis (WEKA) software. Methods: This study was conducted with 508 patients whose SARS-CoV-2 RT-PCR test was positive. These patients were divided into 2 groups, with and without thoracic CT involvement typical for COVID-19. Then, those patients who had signs of typical involvement for COVID-19 in their thoracic CT were divided into 3 groups depending on the extent of their lesions. J48 Decision Tree classification and Linear Regression methods were used on the WEKA software. The codes implemented in the Python programming language were used at the estimation, classification and testing stages. Results: Thoracic CT scans showed that lung involvement was absent in 93 of the patients, mild in 114, moderate in 115, and severe in 159. The success rates of WEKA Linear Regression Formulas calculated using laboratory values and demographic data, respectively 78.92%, 71.69% and 91%. The success rate of the J48 Decision Tree formula used to predict the presence of involvement in thoracic CT was found to be 95.95%. The success rate of the J48 Decision Tree, which was used to predict the degree of involvement in thoracic CT, was 84.39%. The success rate of the J48 Decision Tree used to predict the need for intensive care was found to be 93.06%. Conclusion: The results of this study will facilitate revealing the presence of lung involvement and identification of critical patients in the COVID-19 pandemic and particularly under circumstances and can be used effectively to ensure triage.