Background: The metabolic profile of a patient can changes when Deep Venous Thrombosis (DVT) takes place. It can be used on the disease’s diagnosis and to develop of an artificial intelligence predictive model.
Aims: To obtain a neural network (ANN) to classify patients who have DVT (or not) based on their metabolic profile, obtained by Magnetic Resonance spectroscopy (NMR), and clinical data.
Methods: Table 1 summarizes the clinical patient characteristics. These patients presented a unique confirmed previous DVT of lower limbs or cerebral area and up to three years after the acute episode. Data were collected between 2015 and 2018. The inputs were: sex, age, body mass index, first and second-grade family history of a previous episode, diabetes, arterial hypertension, plus the considered metabolites (signal intensity at the respective chemical shift): lipids (unsaturated fatty acids, triacylglycerides, and glycerides), isoleucine, leucine, valine, lactate, alanine, and glucose. These metabolites were previously determined as important biomarkers using the liquid-state 1H-NMR data acquisition combined with chemometrics analysis. The ANNs had 56 inputs and the output was the DVT outcome. The best ANN was defined by a 3-fold cross-validation scheme. ANNs with one and two hidden layers were tested. The number of neurons ranged from 5 to 35.
|Number of Patients||40||50|
|Age (years)||37.5 (18-78)||40 (23-66)|
|Average Body Mass Index (kg/m²)||25 (19-60)||24 (18-36)|
|Family History of VTE||23||0|
Results: The best result was obtained with a two hidden layer ANN, containing 20 and 5 neurons in the first and second hidden layers, respectively. All the statistical parameters found for the best model are shown in Table 2. The model presented high ACC, AUC, TPR, and TNR. Also, it presented a low probability to indicate false results, which was confirmed by the LR+ and LR- values.
|Accuracy (ACC)||93.3 %|
|Positive Predicted Value (PPV)||0.860|
|Negative Predicted Value (NPV)||1.000|
|True Positive Rate (TPR)||1.000|
|False Positive Rate (FPR)||0.110|
|True Negative Rate (TNR)||0.890|
|Positive Likelihood Ratio (LR+)||9.000|
|Negative Likelihood Ratio (LR-)||0.000|
Conclusions: ANNs have potential to contribute to the diagnosis of DVT. If this model can be validated in another group of patients they can be used by clinicians on daily basis.
To cite this abstract in AMA style:Martins TD, Quintero M, Tasic L, Costa TBBC, Stanisic D, Montalvão SAL, Huber SC, de Paula EV, Filho RM, Annichino-Bizzacchi JM. Diagnosing Deep Venous Thrombosis using Artificial Neural Networks and Metabolomics by 1H-NMR Data [abstract]. Res Pract Thromb Haemost. 2021; 5 (Suppl 2). https://abstracts.isth.org/abstract/diagnosing-deep-venous-thrombosis-using-artificial-neural-networks-and-metabolomics-by-1h-nmr-data/. Accessed November 29, 2023.
« Back to ISTH 2021 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/diagnosing-deep-venous-thrombosis-using-artificial-neural-networks-and-metabolomics-by-1h-nmr-data/