Automated medical diagnosis of COVID-19 through EfficientNet convolutional neural network.

COVID-19 infection was reported in December 2019 at Wuhan, China. This virus critically affects several countries such as the USA, Brazil, India and Italy. Numerous research units are working at their higher level of effort to develop novel methods to prevent and control this pandemic scenario. The main objective of this paper is to propose a medical decision support system using the implementation of a convolutional neural network (CNN). This CNN has been developed using EfficientNet architecture. To the best of the authors' knowledge, there is no similar study that proposes an automated method for COVID-19 diagnosis using EfficientNet. Therefore, the main contribution is to present the results of a CNN developed using EfficientNet and 10-fold stratified cross-validation. This paper presents two main experiments. First, the binary classification results using images from COVID-19 patients and normal patients are shown. Second, the multi-class results using images from COVID-19, pneumonia and normal patients are discussed. The results show average accuracy values for binary and multi-class of 99.62% and 96.70%, respectively. On the one hand, the proposed CNN model using EfficientNet presents an average recall value of 99.63% and 96.69% concerning binary and multi-class, respectively. On the other hand, 99.64% is the average precision value reported by binary classification, and 97.54% is presented in multi-class. Finally, the average F1-score for multi-class is 97.11%, and 99.62% is presented for binary classification. In conclusion, the proposed architecture can provide an automated medical diagnostics system to support healthcare specialists for enhanced decision making during this pandemic scenario.

The Olson method for detection of acute myocardial ischemia in patients with coronary occlusion.

An automated ECG-based method may provide diagnostic support in the management of patients with acute coronary syndrome. The Olson method has previously proved to accurately identify the culprit artery in patients with acute coronary occlusion. METHODS: The Olson method was applied to 360 patients without acute myocardial ischemia and 52 patients with acute coronary occlusion. RESULTS: This study establishes the normal variation of the Olson wall scores in patients without acute myocardial ischemia, which provides the basis for implementation of the Olson method for triage of patients with acute coronary syndrome. All patients with acute occlusion had Olson wall scores above the upper limit of normal. CONCLUSION: The Olson method can be used for ischemia detection with very high sensitivity. Future studies are needed to explore specificity in patients with non-ischemic ST elevation.

Comparison of Insulin Dose Adjustments Made by Artificial Intelligence-Based Decision Support Systems and by Physicians in People with Type 1 Diabetes Using Multiple Daily Injections Therapy.

Objective: Artificial intelligence-based decision support systems (DSS) need to provide decisions that are not inferior to those given by experts in the field. Recommended insulin dose adjustments on the same individual data set were compared among multinational physicians, and with recommendations made by automated Endo.Digital DSS (ED-DSS). Research Design and Methods: This was a noninterventional study surveying 20 physicians from multinational academic centers. The survey included 17 data cases of individuals with type 1 diabetes who are treated with multiple daily insulin injections. Participating physicians were asked to recommend insulin dose adjustments based on glucose and insulin data. Insulin dose adjustments recommendations were compared among physicians and with the automated ED-DSS. The primary endpoints were the percentage of comparison points for which there was agreement on the trend of insulin dose adjustments. Results: The proportion of agreement and disagreement in the direction of insulin dose adjustment among physicians was statistically noninferior to the proportion of agreement and disagreement observed between ED-DSS and physicians for basal rate, carbohydrate-to insulin ratio, and correction factor (P < 0.001 and P ≤ 0.004 for all three parameters for agreement and disagreement, respectively). The ED-DSS magnitude of insulin dose change was consistently lower than that proposed by the physicians. Conclusions: Recommendations for insulin dose adjustments made by automatization did not differ significantly from recommendations given by expert physicians regarding the direction of change. These results highlight the potential utilization of ED-DSS as a useful clinical tool to manage insulin titration and dose adjustments.

Adjustment of Insulin Pump Settings in Type 1 Diabetes Management: Advisor Pro Device Compared to Physicians' Recommendations.

AIMS: To compare insulin dose adjustments made by physicians to those made by an artificial intelligence-based decision support system, the Advisor Pro, in people with type 1 diabetes (T1D) using an insulin pump and self-monitoring blood glucose (SMBG). METHODS: This was a multinational, non-interventional study surveying 17 physicians from 11 countries. Each physician was asked to provide insulin dose adjustments for the settings of the pump including basal rate, carbohydrate-to-insulin ratios (CRs), and correction factors (CFs) for 15 data sets of pumps and SMBG of people with T1D (mean age 18.4 ± 4.8 years; eight females; mean glycated hemoglobin 8.2% ± 1.4% [66 ± 11mmol/mol]). The recommendations were compared among the physicians and between the physicians and the Advisor Pro. The study endpoint was the percentage of comparison points for which there was an agreement on the direction of insulin dose adjustments. RESULTS: The percentage (mean ± SD) of agreement among the physicians on the direction of insulin pump dose adjustments was 51.8% ± 9.2%, 54.2% ± 6.4%, and 49.8% ± 11.6% for the basal, CR, and CF, respectively. The automated recommendations of the Advisor Pro on the direction of insulin dose adjustments were comparable )49.5% ± 6.4%, 55.3% ± 8.7%, and 47.6% ± 14.4% for the basal rate, CR, and CF, respectively( and noninferior to those provided by physicians. The mean absolute difference in magnitude of change between physicians was 17.1% ± 13.1%, 14.6% ± 8.4%, and 23.9% ± 18.6% for the basal, CR, and CF, respectively, and comparable to the Advisor Pro 11.7% ± 9.7%, 10.1% ± 4.5%, and 25.5% ± 19.5%, respectively, significant for basal and CR. CONCLUSIONS: Considerable differences in the recommendations for changes in insulin dosing were observed among physicians. Since automated recommendations by the Advisor Pro were similar to those given by physicians, it could be considered a useful tool to manage T1D.

Origins of the Arden Syntax.

The Arden Syntax originated in the 1980's, when several knowledge-based systems began to show promise, but researchers recognized the burden of recreating these systems at every institution. Derived initially from Health Evaluation through Logical Processing (HELP) and the Regenstrief Medical Record System (RMRS), the Arden Syntax defines medical logic that can be encoded as independent rules, such as reminders and alerts, with the hope of creating a public library of rules. It was first vetted at an informatics retreat held in 1989 at Columbia University's Arden Homestead. The syntax was intended to be readable by clinician experts but to provide powerful array processing, which was derived largely a programming language called APL. The syntax was improved and implemented by a number of researchers and vendors in the early 1990's and was initially adopted by the consensus standards organization, the American Society for Testing and Materials.

Adaptive neuro-heuristic hybrid model for fruit peel defects detection.

Fusion of machine learning methods benefits in decision support systems. A composition of approaches gives a possibility to use the most efficient features composed into one solution. In this article we would like to present an approach to the development of adaptive method based on fusion of proposed novel neural architecture and heuristic search into one co-working solution. We propose a developed neural network architecture that adapts to processed input co-working with heuristic method used to precisely detect areas of interest. Input images are first decomposed into segments. This is to make processing easier, since in smaller images (decomposed segments) developed Adaptive Artificial Neural Network (AANN) processes less information what makes numerical calculations more precise. For each segment a descriptor vector is composed to be presented to the proposed AANN architecture. Evaluation is run adaptively, where the developed AANN adapts to inputs and their features by composed architecture. After evaluation, selected segments are forwarded to heuristic search, which detects areas of interest. As a result the system returns the image with pixels located over peel damages. Presented experimental research results on the developed solution are discussed and compared with other commonly used methods to validate the efficacy and the impact of the proposed fusion in the system structure and training process on classification results.

Referent tracking for treatment optimisation in schizophrenic patients: A case study in applying philosophical ontology to diagnostic algorithms.

The IPAP Schizophrenia Algorithm was originally designed in the form of a flow-chart to help physicians optimise the treatment of schizophrenic patients in the spirit of guideline-based medicine. We take this algorithm as our starting point in investigating how artifacts of this sort can benefit from the facilities of high-quality ontologies. The IPAP algorithm exists thus far only in a form suitable for use by human beings. We draw on the resources of Basic Formal Ontology (BFO) in order to show how such an algorithm can be enhanced in such a way that it can be used in Semantic Web and related applications. We found that BFO provides a framework that is able to capture in a rigorous way all the types of entities represented in the IPAP Schizophrenia Algorithm in way which yields a computational tool that can be used by software agents to perform monitoring and control of schizophrenic patients. We discuss the issues involved in building an application ontology for this purpose, issues which are important for any Semantic Web application in the life science and healthcare domains.