METHODS: A dataset containing medical records of 809 patients suspected to suffer from ACS was used. For each subject, 266 clinical factors were collected. At first, a feature selection was performed based on interviews with 20 cardiologists; thereby 40 seminal features for classifying ACS were selected. Next, a feature selection algorithm was also applied to detect a subset of the features with the best classification accuracy. As a result, the feature numbers considerably reduced to only seven. Lastly, based on the seven selected features, eight various common pattern recognition tools for classification of ACS were used.
RESULTS: The performance of the aforementioned classifiers was compared based on their accuracy computed from their confusion matrices. Among these methods, the multi-layer perceptron showed the best performance with the 83.2% accuracy.
CONCLUSION: The results reveal that an integrated AI-based feature selection and classification approach is an effective method for the early and accurate classification of ACS and ultimately a timely diagnosis and treatment of this disease.
METHODS: Eight internal carotid arteries from different medical centers were diagnosed as stenosed internal carotid arteries, as plaques were found at different locations on the vessel. A computational fluid dynamics solver was developed based on an open-source code (OpenFOAM) to test the flow ratio and energy loss of those stenosed internal carotid arteries. For comparison, a healthy internal carotid artery and an idealized internal carotid artery model have also been tested and compared with stenosed internal carotid artery in terms of flow ratio and energy loss.
RESULTS: We found that at a given common carotid artery bifurcation, there must be a certain flow distribution in the internal carotid artery and external carotid artery, for which the total energy loss at the bifurcation is at a minimum; for a given common carotid artery flow rate, an irregular shaped plaque at the bifurcation constantly resulted in a large value of minimization of energy loss. Thus, minimization of energy loss can be used as an indicator for the estimation of internal carotid artery stenosis.
MATERIALS AND METHODS: Pro-arrhythmic properties in electrocardiographic and intracellular recordings were compared in young and aged, peroxisome proliferator-activated receptor-γ coactivator-1β knockout (Pgc-1β-/-) and wild type (WT), Langendorff-perfused murine hearts, during regular and programmed stimulation (PES), comparing results by two-way ANOVA.
RESULTS AND DISCUSSION: Young and aged Pgc-1β-/- showed higher frequencies and durations of arrhythmic episodes through wider PES coupling-interval ranges than WT. Both young and old, regularly-paced, Pgc-1β-/- hearts showed slowed maximum action potential (AP) upstrokes, (dV/dt)max (∼157 vs. 120-130 V s-1), prolonged AP latencies (by ∼20%) and shortened refractory periods (∼58 vs. 51 ms) but similar AP durations (∼50 ms at 90% recovery) compared to WT. However, Pgc-1β-/- genotype and age each influenced extrasystolic AP latencies during PES. Young and aged WT ventricles displayed distinct, but Pgc-1β-/- ventricles displayed similar dependences of AP latency upon (dV/dt)max resembling aged WT. They also independently increased myocardial fibrosis. AP wavelengths combining activation and recovery terms paralleled contrasting arrhythmic incidences in Pgc-1β-/- and WT hearts. Mitochondrial dysfunction thus causes pro-arrhythmic Pgc-1β-/- phenotypes by altering AP conduction through reducing (dV/dt)max and causing age-dependent fibrotic change.
METHODS: We first tested ten traditional machine learning algorithms, and then the three-best performing algorithms (three types of SVM) were used in the rest of the study. To improve the performance of these algorithms, a data preprocessing with normalization was carried out. Moreover, a genetic algorithm and particle swarm optimization, coupled with stratified 10-fold cross-validation, were used twice: for optimization of classifier parameters and for parallel selection of features.
RESULTS: The presented approach enhanced the performance of all traditional machine learning algorithms used in this study. We also introduced a new optimization technique called N2Genetic optimizer (a new genetic training). Our experiments demonstrated that N2Genetic-nuSVM provided the accuracy of 93.08% and F1-score of 91.51% when predicting CAD outcomes among the patients included in a well-known Z-Alizadeh Sani dataset. These results are competitive and comparable to the best results in the field.
CONCLUSIONS: We showed that machine-learning techniques optimized by the proposed approach, can lead to highly accurate models intended for both clinical and research use.