Knowledge Distillation from General Multi-Objective ECG Classification Model for Chagas Disease Detection in 12-Lead ECGs

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68081731%3A_____%2F25%3A00648258" target="_blank" >RIV/68081731:_____/25:00648258 - isvavai.cz</a>

Result on the web

<a href="https://www.cinc.org/archives/2025/pdf/CinC2025-132.pdf" target="_blank" >https://www.cinc.org/archives/2025/pdf/CinC2025-132.pdf</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.22489/CinC.2025.132" target="_blank" >10.22489/CinC.2025.132</a>

Result language

angličtina

Original language name

Knowledge Distillation from General Multi-Objective ECG Classification Model for Chagas Disease Detection in 12-Lead ECGs

Original language description

We present our solution (team ISIBrno-AIMT) for the 2025 George B. Moody PhysioNet Challenge, which utilizes a teacher-student model architecture. A key component of our approach is a generalized multi-objective teacher model based on the U-Net architecture, which was pre-trained on publicly available 12-lead ECG databases (containing over 1 million ECG recordings) for both the segmentation task and the classification task of 28 cardiac pathologies. The student models were specifically trained to distill knowledge from the teacher’s classification outputs, as well as to predict Chagas disease based on public challenge datasets containing Chagas disease labels. To improve the reliability of our predictions, we employed an ensemble of five student models, averaging their outputs during the inference stage. Our model achieved 4th place in the challenge, with an overall challenge score of 0.271.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

20601 - Medical engineering

Project

<a href="/en/project/FW06010766" target="_blank" >FW06010766: Remote Therapy in Heart Failure Patients Using Artificial Intelligence and Fusion of Multimodal Inputs</a><br>

Continuities

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Article name in the collection

Computing in Cardiology 2025 (CinC 2025)

ISBN

—

ISSN

2325-887X

e-ISSN

—

Number of pages

4

Pages from-to

132

Publisher name

Computing in Cardiology

Place of publication

Neuveden

Event location

Sao Paulo

Event date

Sep 14, 2024

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—