Echo IQ Ltd

Invention:	ML echo analysis
Name:	EchoSolv™
FDA Approval:	Yes (EchoSolv-AS only)
Pathway:	Class II 510(k)
ASX:	EIQ

---

< ALL COMPANIES

THE INVENTION
​
​EchoSolv™ is a cloud-based software platform that applies machine learning to the analysis of echocardiographic data. The suite comprises EchoSolv-AS, designed for the detection of aortic stenosis, and EchoSolv-HF, developed for the diagnosis of heart failure. The EchoSolv model is a Mixture Density Network (MDN) trained on echocardiograms from the National Echocardiographic Database of Australia (NEDA) - the world’s largest mortality-linked echocardiography database. Echo IQ holds an exclusive license with NEDA for model development and training.

A Mixture Density Network (MDN) is a class of neural network that outputs the parameters of a probability distribution rather than a single deterministic value. Unlike image-based systems - which typically employ Convolutional Neural Networks (CNNs) that work directly on scan images - EchoSolv analyses more than 140 quantitative measurements obtained from the echocardiography report itself and generates a heart failure or aortic stenosis risk score in under one minute. While CNNs require complete data to perform reliably, the MDN architecture accommodates missing or noisy inputs through probabilistic modelling and imputation, ensuring consistent performance in real-world clinical settings.

EchoSolv-AS received FDA 510(k) clearance in Q4 2024, while a Mayo Clinic validation study for EchoSolv-HF is currently in progress to support a planned 510(k) submission in Q4 2025.

​VALIDATION STUDIES

EchoSolv-AS
​
Retrospective cohort study (Beth Israel Deaconess)

From an initial cohort of 631,824 individuals, 442,276 were used to train a Mixture Density Network (MDN) designed to predict the echocardiographic phenotype associated with severe aortic stenosis (defined as an aortic valve area < 1.0 cm²). Unlike conventional models, the MDN architecture does not assume a fixed data distribution, allowing it to remain robust to measurement inaccuracies and missing values common in real-world echo data—such as left ventricular outflow tract (LVOT) diameter errors. Importantly, the model was trained not on direct stenosis measures alone, but on the broader physiological “AS phenotype,” including left ventricular dysfunction and hypertrophy, enabling it to generalize beyond perfect data inputs.

The AI required only five core variables—height, weight, sex, left ventricular ejection fraction, and aortic peak velocity—to generate reliable predictions and could scale its accuracy with additional available inputs. When evaluated on a 30% held-out test sample comprising 184,301 individuals, the model achieved an area under the ROC curve (AUC) of 0.986 (95% CI 0.985–0.987) for identifying cases consistent with an AVA < 1.0 cm². This extraordinarily high discrimination demonstrates the capability of EchoSolv-AS to detect severe aortic stenosis from routine echocardiographic parameters with near-perfect accuracy. Sensitivity, specificity, positive predictive value, and negative predictive value (NPV) were 82.2%, 98.1%, 67.4%, and 99.2%, respectively.

​Retrospective study (St Vincent's)
​
​In a retrospective validation study conducted at St Vincent’s Hospital, EchoSolv-AS was applied to 9,189 echocardiograms from the hospital’s database. The model identified 317 patients who met guideline-defined criteria for severe aortic stenosis, encompassing all cases previously recognised by clinicians and revealing an additional 142 cases that had not been detected. In other words, EchoSolv-AS detected 72% more cases of severe AS than standard human assessment alone.
​

EchoSolv-HF
​
The imputation model was trained on 126,136 individual echocardiograms, and another 254,735 scans were used to train the heart failure AI system. A further set of 81,509 echos from NEDA were used for testing. The model produced an output even when traditional guidelines were "indeterminate" (e.g. in the 45.2% of patients with normal ejection fraction but indeterminate filling pressures). The higher the output, the more likely patients were to have the echo findings typically associated with heart failure, and were more likely to die. 

The EchoSolv-HF model was then retrospectively evaluated on 290 participants from the SCREEN-HF trial, comprising 145 patients with clinically confirmed heart failure at follow-up and 145 without. The model achieved an area under the receiver operating characteristic curve (AUC) of 0.86 (95% CI 0.81–0.90; p < 0.001). In probabilistic terms, this means there was an 86% probability that the model assigned a higher heart-failure probability to a randomly selected true case than to a randomly selected control.
​
The model was also evaluated on 453 patients from the NIL-CHF trial, all of whom had a cardiovascular problem other than heart failure, and 93 of whom developed clinical heart failure at three years. In this sample, the model achieved an AUC of 0.89 (95% CI 0.85–0.93; p < 0.001). Crucially, among the 133 patients the AI identified as high-risk at three years, the addition of clinical information (symptoms) identified 97% of patients who went on to develop clinical heart failure or require hospitalization for heart failure within five years.
​
These findings are noteworthy given that, in current clinical practice, heart failure is estimated to be detected in as few as 46% of cases (Sandhu et al.).

FDA 510(k) TRIAL

Echo IQ has initiated a validation study for EchoSolv-HF in collaboration with the Mayo Clinic Platform through the Mayo Validate program—an in-market AI evaluation framework that independently assesses accuracy, efficacy, and bias in clinical software. The study, launched following joint protocol development, will test EchoSolv-HF’s ability to detect heart failure using an independent dataset, generating clinical evidence to support the Company’s forthcoming FDA 510(k) submission.

TOTAL ADDRESSABLE MARKET

The company estimates that the total echo volume in the US that is applicable to EchoSolv-AS is ~5,000,000, and the total applicable to EchoSolv-HF is ~6,000,000. Based on an anticipated reimbursement of US$68 (25% EIQ) for AS and US$284 (25% EIQ) for HF, annual recurring revenue is as follows:

COMMERCIALIZATION

Following FDA clearance, EchoSolv-AS was integrated with Beth Israel Deaconess Medical Center (BIDMC), a leading Harvard Medical School teaching hospital in Boston, Massachusetts. The integration was completed in March 2025. BIDMC is a major academic medical center with 743 beds, approximately 37,600 inpatient discharges, 50,000 emergency visits, 803,000 outpatient visits, and performs around 30,000 echocardiograms annually.

RUNWAY

EchoIQ raised AUD $17.3M via an institutional placement in May 2025. 

REGULATORY MILESTONES

2025

Q4:
Mayo Clinic validation study readout EchoSolv-HF
FDA 510(k) submission EchoSolv-HF
​

SOURCES

Echo IQ ASX Announcement: AI-Backed Solution Accurately Detects Heart Failure in 86% of Patients and 97% of High Risk Individuals: https://announcements.asx.com.au/asxpdf/20240903/pdf/067g3vvr10212f.pdf 

Echo IQ Annual Report 2025: https://api.investi.com.au/api/announcements/eiq/c3451573-87b.pdf
​
Echo IQ Investor Webinar September 12 2024: https://www.youtube.com/watch?v=ON2_VDkA1Vw​

Sandhu, et al. Disparity in the Setting of Incident Heart Failure Diagnosis, PubMed: https://pmc.ncbi.nlm.nih.gov/articles/PMC9070116/
​
Strom, et al. An Artificial Intelligence Algorithm for Detection of Severe Aortic Stenosis - A Clinical Cohort Study: ​https://www.jacc.org/doi/10.1016/j.jacadv.2024.101176