Algorithmic Platform
AI to complement human intelligence with machine capabilities
Thousands of ultrasound images and cardiac cycle clips used for training AI models
Partner with independent physicians for clinical studies
Data partnerships with worldwide clinical institutions
Future Developments - Work In Progress
Kosmos AI Overview
Algorithmic Platform Mission
Kosmos is on a mission to leverage cutting-edge, deep learning algorithms to assist clinicians in assessing heart, lungs and more, with minimal training, in a matter of few minutes.
Innovation At Your Fingertips
Our artificial intelligence platform is developed on convolutional neural networks, which have been trained on thousands of expert annotated ultrasound clips, with an aim to diffuse and scale expert knowledge to frontline care providers.
Accelerate user learning curves and improve confidence with ultrasound-based devices
Provide evidence-based support to help in decision-making
AI-Assisted Ejection Fraction Workflow
Challenge Of Our Times
Patients visiting you with symptoms, but you are unsure of their etiology.
Basic assessment of cardiac functions is important for timely assessment of numerous conditions. However, bedside assessment of key variables such as Ejection Fraction, Stroke Volume, Cardiac Output require considerable skill and experience.
Innovation at Your Fingertips
Our algorithms are developed by applying advanced deep learning techniques, to compute measurements and eliminate the challenges. Furthermore, Kosmos delivers the innovation by emphasizing transparency in its algorithm measurements.
Standardize measurement of EF, SV and CO
Transparent calculation with ability to adjust algorithmic results
Reduce time delays as seen in current echo testing pathways
- KOSMOS torso is placed on the body to capture the heart cycles of the Apical Four Chamber (A4C) in the first 5 seconds.
- The probe is then rotated to capture the heart cycles of the Apical Two Chamber (A2C) view in 5 seconds.
- The probe is removed and then the in-built AI algorithms will calculate key systolic heart function metrics from these acquisitions, almost instantaneously.
Automated Heart Anatomy Labeling
Challenge Of Our Times
Cardiac scanning can be a challenge with a steep learning curve. Novice users can lose orientation during image acquisition and find it difficult to identify anatomical structures in their imaging plane.
AI Innovation Delivered
Our algorithms apply object detection techniques to reliably identify and annotate structures, even in the case of technically difficult patients. Automated annotation provides real-time assistance in obtaining more accurate images, needed for clinical interpretation.
Future Developments - Work In Progress
Fast, accurate identification of cardiac structures
Important cardiac information is accessible to all clinicians
Acts a mentor through real time assistance of anatomical identification
- When the Kosmos Torso is placed with the right orientation on the body, the algorithm labels the Apical 4 Chamber view of the Heart.
- It is then followed by the A5C view, which includes the aorta and the left ventricular outflow tract.
- Next we see the Apical 2 Chamber view and as the probe is continued to be rotated, the labels show an inverted heart on the screen.
Automated Image Grading
Challenge of our times
Good image acquisition depends on the clinician's training, experience, device type and the patient. Today more than ever, we are faced with larger BMI and other patients, that present image acquisition challenges.
AI Innovation Delivered
Our deep learning algorithms provide a real time image quality score, using an established POCUS grading system of 1 to 5. Based on the image appearance and key anatomical structures, the algorithm grades the quality of the acquired images, real time. Once an acceptable image quality is captured, the algorithm then proceeds with the calculation of systolic heart function.
Future Developments - Work In Progress
Real-time determination of image quality during the acquisition process
Standardized grading to guide users towards acceptable, optimal view
Drive accuracy for systolic function assessment
- When the Kosmos Torso is placed with the right orientation on the body, the algorithm grades the quality of the image acquisition, real time.
- Acceptable quality is graded 3 and above, and is indicated in green.
- A grade of 5 is considered as an excellent acquisition. When the probe is moved and the ideal orientation is lost, the scoring grade changes to 2, which means the image acquired, cannot be used for AI calculations.
Automated Acquisition Guidance
Challenge of our times
Medical devices can require considerable training effort and time to learn. When hand-eye coordination and spatial awareness is required, substantial variation in individual skill can occur. This affects diagnostic capability and can be difficult to resolve using one training method for all solutions.
AI Innovation Delivered
Our algorithms provide real-time guidance on probe movement, assisting the user to obtain a diagnostically acceptable image.
Future Developments - Work In Progress
Real time, continuous assistance with Kosmos Torso
Accelerate user learning curves and improve confidence with ultrasound based devices
Personalize training as well as real-time assistance
- When the Kosmos Torso is placed with the right orientation on the body, the guidance algorithm recognizes the 4 chamber view of the heart. When the probe is moved and tilted up, the guidance screen indicates tilt down to rectify the orientation.
- However, if the probe is continued to be tilted down, or moved posteriorly, then the heart will be foreshortened, and the system will indicate tilt up on the screen to rectify it.
- If the probe is rotated counter-clockwise, the guidance algorithm will indicate clockwise on the screen, to rectify the orientation.
Step into the future
Anytime. Anyone. Anywhere. Are you ready for a paradigm shift in the standard of care?