Expert-level localization of fine-grained cardiac structures across the transthoracic echocardiogram

Artificial intelligence models for echocardiography interpret studies without identifying the structures that support their outputs. We show that this spatial grounding does not emerge from the pretraining of current vision-language models. General-purpose and medical vision-language models and an open-vocabulary detector recalled 3% or fewer of cardiac structures and scored below a box that never sees the image. EchoLens, a detector trained on 5,912 expert-annotated structures and 4,219 Doppler signals from 136 patients, recalled 87% and localized structures as reliably as a blinded expert reader (box centre within the reference in 95.7% and 95.3% of cases). It covers 21 structures across 11 transthoracic views and 11 spectral Doppler signal types. In external data, two physicians judged 81-92% of its detections correct across all 11 views, and it identified the Doppler signal type in 83% of frames. EchoLens provides the spatial layer that grounded echocardiographic interpretation requires.

Authors

Institutions

Publication Details

Journal
medRxiv
Published
2026-11-12
DOI
https://doi.org/10.64898/2026.11.10.26367643
Primary Topic
Artificial Intelligence in Healthcare and Education
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Expert-level localization of fine-grained cardiac structures across the transthoracic echocardiogram

Hesham Sheashaa, Bradley James Erickson, Elham Mahmoudi, Chadi Ayoub et al.
medRxiv
Artificial Intelligence in Healthcare and Education
preprint

Expert-level localization of fine-grained cardiac structures across the transthoracic echocardiogram

Hesham Sheashaa, Bradley James Erickson, Elham Mahmoudi, Chadi Ayoub, Nicolò Pecco, Imon Banerjee, Chieh‐Ju Chao, Reza Arsanjani, Isabel G. Scalia, Kamal Awad, Martha A. Grogan, Juan Farina, Fatmaelzahraa E. Abdelfattah, Nima Baba Ali, Sogol Attaripour Esfahani, Garvan Kane, Jae K. Oh
preprint en

Abstract

Artificial intelligence models for echocardiography interpret studies without identifying the structures that support their outputs. We show that this spatial grounding does not emerge from the pretraining of current vision-language models. General-purpose and medical vision-language models and an open-vocabulary detector recalled 3% or fewer of cardiac structures and scored below a box that never sees the image. EchoLens, a detector trained on 5,912 expert-annotated structures and 4,219 Doppler signals from 136 patients, recalled 87% and localized structures as reliably as a blinded expert reader (box centre within the reference in 95.7% and 95.3% of cases). It covers 21 structures across 11 transthoracic views and 11 spectral Doppler signal types. In external data, two physicians judged 81-92% of its detections correct across all 11 views, and it identified the Doppler signal type in 83% of frames. EchoLens provides the spatial layer that grounded echocardiographic interpretation requires.

medRxiv
Cleveland Clinic (US), Mayo Clinic (US), Vita-Salute San Raffaele University (IT), Mayo Clinic Hospital (US)
Good health and well-being
Artificial Intelligence in Healthcare and Education
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon