Background: Deep vein thrombosis (DVT) is a common multifactorial disease. If the disease is not prevented as soon as possible, it may prolong the cure time and even lead to death.

Aims: To develop a simple, cost-effective, accurate and non-invasive prediction model by artificial intelligence machine-learning model for early detection of high-risk DVT patients.

Methods: 1295 patients were selected including all 729 patients with

DVT and 566 patients without from January 2011 to December 2020 in the 6th

Medical Centre of Chinese PLA General Hospital.  An artificial neural network (ANN) predictive procedure divided three-quarters of the patients (993 cases) randomly as training sample data to construct the predictive model while one quarter of the

patients (302 cases) used to test the effectiveness of the constructed model. ANN inputs were gender, age, fibrinogen, D-Dimer, hypertension, coronary heart disease, tumor,diabetes, chemotherapy, surgery etc.

Results: When the number of hidden-nodes was set to 8 and the number of epochs was set to 800 in the artificial neural network model, the accuracy reached the

highest with an average rate of 81.47% and variance of 0.0001. The prediction

accuracy of ANN model was higher than that of linear Logistic regression model

with an average rate of 79.80%.

Conclusion(s): With the help of computer-based ANN model, the high-risk DVT

patients could be detected as early as possible, conducive to disease prevention and the DVT incidence reduction.

« Back to ISTH 2022 Congress

ISTH Congress Abstracts - https://abstracts.isth.org/abstract/predicting-dvt-incidence-based-on-risk-factors-by-using-artificial-neural-networks/