The editor of the Journal of NeuroTechnology prof. Francesco Carlino will attend the:
ESC CONGRESS 2015 (Saturday 29 August - Wednesday 02 September 2015 (5 Days) , London - United Kingdom)
with the following speech:
Title : Artificial intelligence in cardiac imaging. applications on hand held echo. preliminary data.
Topic : 09.15 - Image databases, image data transmission, image analysis
Category : Bedside
Option : No Options
M. Mazzanti1, F. Carlino2 - , Italy
Introduction. Hand held echo (HHE) highly entered the market and its non-cardiologists clinical use at point-of-care (POC) is increasing. Nuclear Cardiology, CT/MRI showed that artificial intelligence (AI) learning will likely assist physicians with cues in automatic tool for measurements calculation. Differently, echocardiography was focused several times but any application entered today in routine use on HHE. Purpose. Aim of the study was to develop and clinically use a software tool (ST) which automatically calculate the dimensions (D) and the systolic function (SF) of the left ventricle (LV) from HHE images suggesting interpretation of the results. Methods. A ST architecture has been implemented including the synergy between two types of AI neural networks (NN). In the first level of analysis, a Self-Organizing Maps (SOM) NN (classifiable as a generalization of a Kohonen Neural Network-KNN:unsupervised network-UN) has the purpose of distinguishing the noise of the echo image from the presence of cardiac tissue. Therefore the UN has the goal of highly precision segmenting LV myocardium HHE images. In the second phase of analysis, a UN (classified as the Error Back Propagation - EBP family) is trained to recognize the morphology of LV myocardium (M). This phase has a dual purpose: a) showing to the operator while positioning the echo probe that optimal window to perform the bio-medical measurements (BM) has been reached; b) automatically and instantly carrying out LV D and LV SF dynamic. 21 subjects with very low likelihood of having coronary artery disease (CAD), 13 males and 8 females, age 53 ± 5 years with no hypertension performed resting HHE imaging with Broad-bandwidth phased array probe from 1.7 to 3.8 MHz applying SOM (KNN+EBP) automatically on apical-4- chambers views LV M. Automatic identification of endocardial and epicardial borders provides LV D, LV EndDiastolic (LVEDV), EndSystolic Volumes (LVESV) and LV Ejection Fraction (LVEF). Results. In 19 of 21 subjects SOMs were able to identify the optimal window for operator to perform acquisition recording . The contour of LV M was automatically well traced in 95% of subjects (20 subjects). LV D resulted in 44±6 mm; LVEDV of 44±6 mL/m2; LVESV of 17±4 mL/m2 and LVEF of 0,59±0,09. Conclusions. AI NN ST is able to identify and make the contour of LV M. Our preliminary data demonstrates that AI ST optimally and automatically identifies the window for performing calculations and showing BM data. This represent an encouraging promise in clinical applications expecially for regional wall motion and thickness analysis at POC.