CORTEXA
← Browse
zenodoConference paper2026-02-15

A Modular Deep Learning Pipeline for Echocardiographic Video Classification of Heart Failure

Tijana Geroski, Mahyar Bolhassani, Narayana S. Singam, Ayman Battisha, Nenad Filipović, Dinesh Kalra, Amir A. Amini

Distinguishing heart failure with reduced ejection fraction (HFrEF) from heart failure with preserved ejection fraction (HFpEF) is clinically important but challenging. This paper presents an end-to-end deep learning pipeline for automated three-class classification (healthy, HFrEF, HFpEF) from 2D apical four-chamber echocardiographic videos, combining the public RVENet dataset with institutional data from the University of Louisville. After a multi-step preprocessing pipeline (colour-space conversion, cardiac-cycle detection, motion-based cropping, normalization), a lightweight 3D SqueezeNet and a deeper ResNet50 were evaluated as spatio-temporal classifiers, achieving AUC values of 0.93-0.98 across classes, with ResNet50 reaching an F1-score of up to 0.9205 for HFrEF when using two cardiac cycles as input. This work was presented at SPIE Medical Imaging 2026: Clinical and Biomedical Imaging, Vancouver, BC, Canada, and was carried out in part within the STRATIFYHF project.

View free PDFSource page

Related papers

zenodoConference paper2026-07-28

KeepA(n)I: Social Stereotypes in and Social Norms for Computer Vision

Evgenia Christoforou, Nicolas Nicolaou, Efstathios Stavrakis, Jahna Otterbacher

The KeepA(n)I platform facilitates the auditing of computer vision systems that tag images, which aid visual communication on the Web and social media, from content moderation to the development of new apps and tools. In particular, KeepA(n)I enables a broad set of stakeholders t…

View free PDFSource page
zenodoConference paper2026-07-28

Self-Supervised Relevance Modelling in Autonomous Driving via Counterfactual Analysis

Luca Lusvarghi, Javier Gozalvez, Pablo Urbano Hidalgo

Autonomous driving relies on computationally intensive perception pipelines to continuously detect and track objects in the surrounding environment. While some objects are key to plan safe and effective maneuvers, others may not be relevant and have no impact on the autonomous ve…

View free PDFSource page
zenodoConference paper2026-05-27

Обмеження навчання моделей комп'ютерного зору для систем розпізнавання психоемоційних станів в авіаційній інфраструктурі

Наталія Подопригора

У статті досліджуються епістемологічні та алгоритмічні обмеження сучасних моделей комп'ютерного зору (зокрема архітектури YOLOv8) щодо автоматизованого розпізнавання складних психоемоційних станів в умовах авіаційної інфраструктури. На основі трансдисциплінарної методології…

View free PDFSource page
zenodoConference paper2026-05-20

Differentiating Suspected and Confirmed Heart Failure Using Machine Learning and Refined Vocal Features

Lazar Dašić, Ognjen Pavić, Tijana Geroski, Anđela Blagojević, Andrej Preveden, Aleksandra Milovančev, et al.

Voice characteristics are an emerging, non-invasive biomarker for heart failure. This study develops a machine learning pipeline to differentiate patients with suspected heart failure from those with a confirmed diagnosis using vocal features alone, drawing on 240 patients (50 su…

View free PDFSource page
zenodoConference paper2026-05-20

SoC-Based Implementation of CNN Model for End-Diastolic Volume Classification from Echocardiogram via hls4ml

Nemanja Marković, Tijana Geroski, Emil Jovanov, Nenad Filipović

Echocardiographic assessment of End-Diastolic Volume (EDV) is central to identifying dilated cardiomyopathy, a major driver of heart failure. This paper presents a low-latency, edge-computing solution that deploys an 8-bit quantized, 50%-pruned CNN directly onto a Xilinx Artix-7…

View free PDFSource page