Artificial neural network-based model enhances risk stratification and reduces non-invasive cardiac stress imaging compared to Diamond-Forrester and Morise risk assessment models: A prospective study.

BACKGROUND: Coronary artery disease (CAD) accounts for more than half of all cardiovascular events. Stress testing remains the cornerstone for non-invasive assessment of patients with possible or known CAD. Clinical utilization reviews show that most patients presenting for evaluation of stable CAD by stress testing are categorized as low risk prior to the test. Attempts to enhance risk stratification of individuals who are sent for stress testing seem to be more in need today. The present study compares artificial neural networks (ANN)-based prediction models to the other risk models being used in practice (the Diamond-Forrester and the Morise models). METHODS: In our study, we prospectively recruited patients who were 19 years of age or older, and were being evaluated for coronary artery disease with imaging-based stress tests. For ANN, the network architecture employed a systematic method, where the number of neurons is changed incrementally, and bootstrapping was performed to evaluate the accuracy of the models. RESULTS: We prospectively enrolled 486 patients. The mean age of patients undergoing stress test was 55.2 ± 11.2 years, 35% were women, and 12% had a positive stress test for ischemic heart disease. When compared to Diamond-Forrester and Morise risk models, the ANN model for predicting ischemia provided higher discriminatory power (DP)(1.61), had a negative predictive value of 98%, Sensitivity 91% [81%-97%], Specificity 65% [60%-79%], positive predictive value 26%, and a potential 59% reduction of non-invasive imaging. CONCLUSION: The ANN models improved risk stratification when compared to the other risk scores (Diamond-Forrester and Morise) with a 98% negative predictive value and a significant potential reduction in non-invasive imaging tests.

Improved accuracy of anticoagulant dose prediction using a pharmacogenetic and artificial neural network-based method.

BACKGROUND: The unpredictability of acenocoumarol dose needed to achieve target blood thinning level remains a challenge. We aimed to apply and compare a pharmacogenetic least-squares model (LSM) and artificial neural network (ANN) models for predictions of acenocoumarol dosing. METHODS: LSM and ANN models were used to analyze previously collected data on 174 participants (mean age: 67.45 SD 13.49 years) on acenocoumarol maintenance therapy. The models were based on demographics, lifestyle habits, concomitant diseases, medication intake, target INR, and genotyping results for CYP2C9 and VKORC1. LSM versus ANN performance comparisons were done by two methods: by randomly splitting the data as 50 % derivation and 50 % validation cohort followed by a bootstrap of 200 iterations, and by a 10-fold leave-one-out cross-validation technique. RESULTS: The ANN-based pharmacogenetic model provided higher accuracy and larger R value than all other LSM-based models. The accuracy percentage improvement ranged between 5 % and 24 % for the derivation cohort and between 12 % and 25 % for the validation cohort. The increase in R value ranged between 6 % and 31 % for the derivation cohort and between 2 % and 31 % for the validation cohort. ANN increased the percentage of accurately dosed subjects (mean absolute error ≤1 mg/week) by 14.1 %, reduced the percentage of mis-dosed subjects (mean absolute error 2-3 mg/week) by 7.04 %, and reduced the percentage of grossly mis-dosed subjects (mean absolute error ≥4 mg/week) by 24 %. CONCLUSIONS: ANN-based pharmacogenetic guidance of acenocoumarol dosing reduces the error in dosing to achieve target INR. These results need to be ascertained in a prospective study.

Semantic Segmentation of Maxillary Teeth and Palatal Rugae in Two-Dimensional Images.

The superimposition of sequential radiographs of the head is commonly used to determine the amount and direction of orthodontic tooth movement. A harmless method includes the timely unlimited superimposition on the relatively stable palatal rugae, but the method is performed manually and, if automated, relies on the best fit of surfaces, not only rugal structures. In the first step, motion estimation requires segmenting and detecting the location of teeth and rugae at any time during the orthodontic intervention. Aim: to develop a process of tooth segmentation that eliminates all manual steps to achieve an autonomous system of assessment of the dentition. Methods: A dataset of 797 occlusal views from photographs of teeth was created. The photographs were manually semantically segmented and labeled. Machine learning methods were applied to identify a robust deep network architecture able to semantically segment teeth in unseen photographs. Using well-defined metrics such as accuracy, precision, and the average mean intersection over union (mIoU), four network architectures were tested: MobileUnet, AdapNet, DenseNet, and SegNet. The robustness of the trained network was additionally tested on a set of 47 image pairs of patients before and after orthodontic treatment. Results: SegNet was the most accurate network, producing 95.19% accuracy and an average mIoU value of 86.66% for the main sample and 86.2% for pre- and post-treatment images. Conclusions: Four architectural tests were developed for automated individual teeth segmentation and detection in two-dimensional photos that required no post-processing. Accuracy and robustness were best achieved with SegNet. Further research should focus on clinical applications and 3D system development.

Deep learning applications in pulmonary medical imaging: recent updates and insights on COVID-19.

Shortly after deep learning algorithms were applied to Image Analysis, and more importantly to medical imaging, their applications increased significantly to become a trend. Likewise, deep learning applications (DL) on pulmonary medical images emerged to achieve remarkable advances leading to promising clinical trials. Yet, coronavirus can be the real trigger to open the route for fast integration of DL in hospitals and medical centers. This paper reviews the development of deep learning applications in medical image analysis targeting pulmonary imaging and giving insights of contributions to COVID-19. It covers more than 160 contributions and surveys in this field, all issued between February 2017 and May 2020 inclusively, highlighting various deep learning tasks such as classification, segmentation, and detection, as well as different pulmonary pathologies like airway diseases, lung cancer, COVID-19 and other infections. It summarizes and discusses the current state-of-the-art approaches in this research domain, highlighting the challenges, especially with COVID-19 pandemic current situation.

Artificial neural network modeling enhances risk stratification and can reduce downstream testing for patients with suspected acute coronary syndromes, negative cardiac biomarkers, and normal ECGs.

Despite uncertain yield, guidelines endorse routine stress myocardial perfusion imaging (MPI) for patients with suspected acute coronary syndromes, unremarkable serial electrocardiograms, and negative troponin measurements. In these patients, outcome prediction and risk stratification models could spare unnecessary testing. This study therefore investigated the use of artificial neural networks (ANN) to improve risk stratification and prediction of MPI and angiographic results. We retrospectively identified 5354 consecutive patients referred from the emergency department for rest-stress MPI after serial negative troponins and normal ECGs. Patients were risk stratified according to thrombolysis in myocardial infarction (TIMI) scores, ischemia was defined as >5 % reversible perfusion defect, and obstructive coronary artery disease was defined as >50 % angiographic obstruction. For ANN, the network architecture employed a systematic method where the number of neurons is changed incrementally, and bootstrapping was performed to evaluate the accuracy of the models. Compared to TIMI scores, ANN models provided improved discriminatory power. With regards to MPI, an ANN model could reduce testing by 59 % and maintain a 96 % negative predictive value (NPV) for ruling out ischemia. Application of an ANN model could also avoid 73 % of invasive coronary angiograms while maintaining a 98 % NPV for detecting obstructive CAD. An online calculator for clinical use was created using these models. The ANN models improved risk stratification when compared to the TIMI score. Our calculator could also reduce downstream testing while maintaining an excellent NPV, though further study is needed before the calculator can be used clinically.

Artificial neural network modeling using clinical and knowledge independent variables predicts salt intake reduction behavior.

BACKGROUND: High dietary salt intake is directly linked to hypertension and cardiovascular diseases (CVDs). Predicting behaviors regarding salt intake habits is vital to guide interventions and increase their effectiveness. We aim to compare the accuracy of an artificial neural network (ANN) based tool that predicts behavior from key knowledge questions along with clinical data in a high cardiovascular risk cohort relative to the least square models (LSM) method. METHODS: We collected knowledge, attitude and behavior data on 115 patients. A behavior score was calculated to classify patients' behavior towards reducing salt intake. Accuracy comparison between ANN and regression analysis was calculated using the bootstrap technique with 200 iterations. RESULTS: Starting from a 69-item questionnaire, a reduced model was developed and included eight knowledge items found to result in the highest accuracy of 62% CI (58-67%). The best prediction accuracy in the full and reduced models was attained by ANN at 66% and 62%, respectively, compared to full and reduced LSM at 40% and 34%, respectively. The average relative increase in accuracy over all in the full and reduced models is 82% and 102%, respectively. CONCLUSIONS: Using ANN modeling, we can predict salt reduction behaviors with 66% accuracy. The statistical model has been implemented in an online calculator and can be used in clinics to estimate the patient's behavior. This will help implementation in future research to further prove clinical utility of this tool to guide therapeutic salt reduction interventions in high cardiovascular risk individuals.