19 November 2026 · 12 min

Company Spotlight: Us2.ai Us2.ca - AI Detection for Cardiac Amyloidosis

CardiologyRadiology & ImagingFDA Clearances

We unpack the FDA 510(k) summary for Us2.ca, an AI software device by Us2.ai designed to detect cardiac amyloidosis from echocardiograms. We break down the clinical testing data, demographic performance differences, and key limitations hospital buyers need to know.

Key points

Source: FDA 510(k) summary K250151: Us2.ca (Eko.Ai Pte) - U.S. Food and Drug Administration, 2025

This episode is an AI-generated conversation summarising a public document; the hosts' voices are synthetic. It is for information only and is not medical advice. Always refer to the original source.

Company: Eko.Ai Pte · Product: Us2.ca · FDA 510(k) K250151

Your company here. This podcast is looking for its first sponsors: reach clinicians, health-system leaders and medtech and pharma teams following AI in medicine. Sponsorship options and rates →

Transcript

Sam: This AI tool just cleared the FDA by hitting an 86.9% sensitivity for detecting a complex heart condition from standard ultrasounds, so we are looking at the public FDA 510(k) summary for a software device called Us2.ca, published by the U.S. Food and Drug Administration on June 20, 2025.

Maya: And right up front, we want to remind you that our voices are AI-generated, and this episode is a summary of a publicly available regulatory document, not an advertisement or company PR.

Sam: Exactly. We are focusing strictly on what is in this specific FDA filing for Us2.ca, which is manufactured by Eko.ai Pte. Ltd., doing business as Us2.ai. Maya, this is an interesting one for both clinicians and health-system leaders because it touches on AI's role in cardiac imaging. What exactly is this device?

Maya: Us2.ca is a software device, specifically an enhancement to the company's existing Us2.v2 software. It is an image post-processing analysis software used for viewing and quantifying cardiovascular ultrasound images in DICOM format. The summary states its primary intended function is to automatically identify patients who require additional follow-up for cardiac amyloidosis.

Sam: Cardiac amyloidosis. And the document gives very specific parameters for who this software can be used on, right?

Maya: Yes, the indications for use are very precise. It is intended for use only in adult patients with increased left ventricular wall thickness. And it defines that exactly as an interventricular septal thickness, or IVSd, or left ventricular posterior wall thickness, LVPWd, of ≥ 12mm.

Sam: So a clinician or sonographer has to already know the patient has that ≥ 12mm wall thickness before applying this specific AI tool?

Maya: According to the stated indications for use, yes. It processes acquired transthoracic cardiac ultrasound images to support qualified cardiologists, sonographers, or other licensed professional healthcare practitioners.

Sam: Let us talk about how it actually works. The summary describes it as using a video-based deep-learning approach. It analyzes what they call Apical 4-Chamber, or A4C, clips to identify the presence of cardiac amyloidosis. But before we get into the testing numbers, we should explain how it got cleared. This is a 510(k) summary, which means it is based on something called substantial equivalence.

Maya: Right. In the FDA framework, if your device is substantially equivalent to a legally marketed predicate device, you can market it without needing the longer premarket approval process. This falls under the regulation name Adjunctive Cardiovascular Status Indicator, which is a Class II device with the product code SDJ.

Sam: And the predicate device they compared it to is the EchoGo Cardiac Amyloidosis 1.0, manufactured by Ultromics Limited, which received its 510(k) clearance on Nov 15, 2024. The document includes a comparison table between the two.

Maya: It does. Both are software as a medical device, or SaMD, both are machine learning-based algorithms taking DICOM files containing echocardiograms as input, and both target cardiac amyloidosis. There is a slight difference in the intended population, though. The predicate, EchoGo, is indicated for adults over the age of 65 with heart failure, whereas Us2.ca is just indicated for adults generally.

Sam: That is a notable expansion in the target age group. Let us dive into the performance data, because this is where buyers and clinicians really need to pay attention. How did they test this?

Maya: The summary notes it was developed using a training dataset of 4,371 patients. Then, it was externally validated on an independent pooled dataset of 1,647 patients. They stress that the training and external validation datasets came from entirely separate data providers, so there was no overlap.

Sam: And the validation study was a retrospective case-control study. That is an important detail for clinicians evaluating the evidence. Retrospective means they looked back at past echocardiographic studies obtained from routine clinical evaluations, rather than running a prospective trial in real-time.

Maya: Exactly. The data was collected from six clinical sites across the United States and Japan. So out of that testing data of 1,647 patients, there were 664 cardiac amyloidosis cases and 983 control cases.

Sam: Okay, so what happened when they ran those 1,647 patients through the Us2.ca software?

Maya: Here is where the concept of yield comes in. The document states that the software was able to generate a confident prediction for 1434 patients. That is 574 in the cardiac amyloidosis group and 860 in the controls group. So the overall yield was 87.1%.

Sam: Wait, so for the rest of the patients, it could not give a confident prediction? Does the summary explain why?

Maya: It does not give a specific reason for those exact cases in the testing section, but in the limitations section, it warns that inappropriate use of the echo machine, use of non-cardiac ultrasound probes, suboptimal settings like gain, contrast, or depth, or lack of electrocardiogram capture may lead to lower accuracy.

Sam: That makes sense in a real-world workflow. But for the 1434 patients where it did make a confident prediction, the numbers are strong. It performed with a sensitivity of 86.9% and a specificity of 87.4%.

Maya: Yes. And they provide a 95% confidence interval for both. The sensitivity interval is 84.2% to 89.7%, and the specificity interval is 85.2% to 89.7%.

Sam: They also ran subgroup analyses across age, gender, race, ultrasound vendor, and clinical site. This is fascinating to look at. For instance, age. For patients under 65 years, the sensitivity was 78.02%. But for patients 65 years and older, the sensitivity jumped to 88.7%.

Maya: And the specificity flipped slightly. It was 88.63% for the younger group and 86.58% for the 65 and older group. There were also variations by gender. Sensitivity for females was 85.61%, and for males, it was 87.41%.

Sam: The breakdown by race is really detailed too. Asian sensitivity was 92.31%, while African American was 85.38%, and White was 83.7%.

Maya: There was also a subgroup for Hispanic or Latino with a 100% sensitivity, but the demographic table earlier in the document shows there were only 5 Hispanic or Latino patients in the cardiac amyloidosis group and 0 in the control group. So the sample size there is extremely small.

Sam: Good catch. That is why looking at the raw numbers in these regulatory summaries is so critical for health systems evaluating AI. You have to know what the algorithm actually trained and tested on. Speaking of testing, there is a major limitation regarding the ultrasound hardware vendors.

Maya: Yes, a very clear warning. The summary states that ultrasound data were obtained from GE, Philips, and Siemens systems to train the device, but the testing data were only obtained from GE and Philips systems.

Sam: And it explicitly says in the limitations that use of the device on ultrasound systems other than GE and Philips may lead to lower accuracy of the software.

Maya: When you look at the vendor subgroup analysis, there is a difference even between those two. GE had a sensitivity of 93.14% and a specificity of 86.41%. Philips had a sensitivity of 84.17% and a specificity of 88.42%.

Sam: So if you are a hospital IT or clinical leader looking to buy this, you need to look at your fleet of echo machines. If you are entirely on Siemens or another brand, the document plainly states that using this software might lead to lower accuracy.

Maya: That is exactly the kind of boardroom conversation this document informs. And for the clinicians on the ward, there are other critical limitations to understand. The document explicitly states that Us2.ca does not identify regions of interest or areas of possible pathology in the images.

Sam: Wait, really? So it does not highlight the specific part of the heart wall that looks problematic on the screen?

Maya: That is correct. It generates a confident prediction, but it does not point out the visual evidence on the scan. The summary also notes that the software does not differentiate different types of cardiac amyloidosis, giving the examples of hereditary ATTR, wild-type ATTR, and AL.

Sam: So it serves purely as an alert system, basically saying this patient needs further follow-up for cardiac amyloidosis broadly, but the physician has to take it from there.

Maya: Exactly. The summary emphasizes that the software complements good patient care and does not exempt the user from the responsibility to provide supervision, clinically review the patient, and make appropriate clinical decisions. It says multiple times that it is not intended to provide a diagnosis and does not replace current standards of care.

Sam: And it mentions the primary benefit is to improve clinical echocardiographic workflow, enabling clinicians to generate and edit reports faster, with precision and with full control. But the final clinical decision always remains with the clinicians.

Maya: Right. Because this is a Class II device under the Adjunctive Cardiovascular Status Indicator regulation, there are special controls it has to fulfill. The FDA requires things like a usability assessment to demonstrate that the risk of misinterpretation is appropriately mitigated.

Sam: The summary mentions they performed a Human Factors/Usability Engineering study. They also had to do a comprehensive hazard analysis, verification of the algorithm calculations, and data security testing.

Maya: Yes, cybersecurity is specifically listed. The document notes that a Cybersecurity Analysis and data security testing were conducted to verify that data and patient protected health information security measures are included in the design.

Sam: One quick thing I noticed in the data source subgroup table. The text mentions the data was collected from six clinical sites across the United States and Japan, but the subgroup table for data source only lists Center 1, Center 2, Center 3, and Center 4.

Maya: That is true. The table only breaks down the primary analyses for four centers. Center 4 had the highest sensitivity at 92.26%, while Center 2 had a sensitivity of 78.21% but a specificity of 92.11%. The document does not explain why only four centers are in that specific table when the text mentions six.

Sam: Just another reason to read these FDA summaries closely. So, pulling it all together, what is the single most memorable takeaway for our listeners?

Maya: The big takeaway is that Us2.ca offers a high-yield, 87.1% confident prediction rate for detecting cardiac amyloidosis in patients with a wall thickness of ≥ 12mm, achieving an 86.9% sensitivity. However, hospital buyers must remember it was only validated on GE and Philips systems, and it does not show clinicians the region of interest on the image.

Sam: A perfect wrap-up. To recap, we unpacked the FDA 510(k) summary for Us2.ca, a software device for analyzing echocardiogram images to support the diagnosis of cardiac amyloidosis. We covered its performance stats, its demographic breakdowns, and its explicit limitations.

Maya: You can find a link to the original FDA document in our show notes. And as always, this podcast is for informational purposes only and is not medical advice.