A machine learning model to classify aortic dissection patients in the early diagnosis phase

Da Huo,Ming Lv,Zhili Zhou,Bo Kou

doi:10.1038/s41598-019-39066-9

Abstract

Aortic dissection is one of the most clinical-challenging and life-threatening cardiovascular diseases associated with high morbidity and mortality. Aortic dissection requires fast diagnosis and timely therapy. Any delay or misdiagnosis can cause severe consequence to aortic dissection patients with even higher mortality. To better help physicians identify the potential dissection within the scope of all misdiagnosed patients, this paper describes a method which is developed with data mining methods for aortic dissection patient classification and prediction in the phase of early diagnosis. Various machine learning algorithms were used to build the models which were all trained and tested on the patient dataset with cross validation. Among them, Bayesian Network model achieved the best performance by predicting at a precision rate of 84.55% with Area Under the Curve (AUC) value of 0.857. On this basis, the Bayesian Network model can help physicians better with early diagnosis of aortic dissection in clinical practice. Beyond this study, more data from diverse regions and the internal pathology can be crucial to further build a universal model with broader predictive power.

Highlights

Cardiovascular diseases are among the major causes of death in most developed countries and many if not all developing countries[1]
Some works from the previous literature show that patients with aortic dissection share pathologic similarities with other diseases4–6. which may not be so useful to distinguish the real ones from all misdiagnosed cases
Aortic dissection patients usually arrive in an emergency with the common abrupt onset of severe chest pain, which makes it critical and challenging to successfully predict and identify aortic dissection as the actual cause out of all the possible ones[7]

Summary

Introduction

Cardiovascular diseases are among the major causes of death in most developed countries and many if not all developing countries[1]. Beyond traditional detection schema of processing ECG and chest X-ray which is slow that may delay the timely therapy, a new method that can assist physicians to fast and reliably identify the aortic dissection patients from all misdiagnosed cases is badly in need. In order to fast and reliably identify, or in other words, to classify patients into aortic dissection category if they are, it is necessary to conduct methods which have high effectiveness and efficiency in classifying In view of this strong demand in clinical practice, machine learning classification can serve as one solution. We proposed a prediction model by applying classification analysis to classify aortic dissection positive patients from all misdiagnosed cases in the phase of early diagnosis. It saves both patients’ time and hospital resources by making it possible to provide the timely therapy for the most needed patients

Methods

Results

Conclusion