Technology Challenges and Opportunities for Obstetric Ultrasound
Obstetric ultrasound is not routinely available in LMICs, particularly at the PHC level where pregnant women first seek care. The primary challenges are to make it more affordable and less dependent on specialized skills. The cost of a portable ultrasound device with a handheld probe can be reduced by engaging existing device manufacturers. By incorporating AI-based software for data analysis, non-sonographer care workers can acquire key obstetric information through “blind sweeps” with the ultrasound probe (i.e., unguided and without real-time feedback). Such a tool would be used not for definitive diagnosis, but to support clinical decision making, including triage and referral. Innovation is needed, however, for both the hardware and software components and for their effective integration and use in low-resource settings. GH Labs supports the development of AI-based models for POCUS devices.
Screening
An affordable, AI-supported point-of-care ultrasound (POCUS) device could identify risk factors for a healthy pregnancy and birth as a routine part of antenatal care. Accurate AI-based algorithms could determine, for instance, gestational age, the presence of more than one fetus, and whether the fetus is in a breech orientation. AI models enabling ultrasound imaging through “blind sweeps” could significantly lower the barrier for use by health care workers without extensive ultrasound training, e.g., nurses and midwives. Effective implementation of AI-supported POCUS devices at the PHC level could also enable their use across an expanded set of health conditions and patients, obstetric and beyond. Regulatory guidance, such as that by the U.S. FDA for ultrasound devices and medical imaging software, supports companies in innovating and expanding to new use cases.
GH Labs supports algorithm development for several key features and feature identification (gestational age, fetal presentation, estimated fetal weight, multiple gestation, amniotic fluid volume). These AI models are directly shared with multiple partners (Edan and GE Healthcare), enabling multiple partners to create technologies for the market.
Point-of-Care Ultrasound (POCUS) Hardware
A variety of innovations are making ultrasound devices more portable, affordable, usable without specialized training, and accessible for broad use in LMICs. There are existing low-cost device manufacturers, including some in China. There are also existing companies producing components for ultrasound devices across a range of medical uses. This includes different ultrasound probes enabling individual devices to be used for multiple health conditions, obstetric as well as others, such as cardiac and pulmonary conditions. (Ultrasound probes contain transducers that convert electric signals to outgoing sound waves and convert incoming sound waves to electrical signals. Transducers can be arrayed in different spatial configurations in the probe, including a convex, linear, or phased array; these arrays confer different capabilities, with convex arrays preferred for obstetrics because they confer a wider field of view useful for imaging deeper structures, such as the fetus during pregnancy. There are also different types of transducers: piezoelectric transducers (PZT) are the most common in existing commercial devices; piezoelectric micromachined ultrasonic transducers (PMUT) and capacitive micromachined ultrasonic transducers (CMUT) are newer and are being used to facilitate miniaturization, improve integration with electronics, and have a wide bandwidth enabling a one-probe-solution for AI multiple use cases.)
In the table below, technologies are assessed across commercial and technical dimensions. To address commercial challenges associated with screening in LMICs, these screening devices should have regulatory approval from WHO recognized stringent regulatory authorities (e.g., FDA, CE Mark) or country level regulatory bodies and have a low price per device that countries can procure for PHCs. A company’s commercial presence and experience in LMICs can help accelerate product introduction, and the volume of devices indicates their market penetration. Technically, ultrasounds require hardware that can produce high quality images; thus, transducers leverage different technology to enhance their usability. Furthermore, AI-based software can be added to ultrasounds to guide technicians on image capture, provide feedback on image quality, and support clinical interpretation of ultrasound images, increasing accuracy and addressing limitations of skilled workers at PHCs.
The table provides an overview of the state of innovation, with the listed technologies serving as representative samples of products available in the market or in development.
Name | Innovator | Description | Image | Development Status (Release Year) + Regulatory Approval (Approval Year) | Price per Device | LMIC Experience | Volume Sold | Transducer | AI-based software | Learn more |
|---|---|---|---|---|---|---|---|---|---|---|
Edan | Wireless handheld color doppler ultrasound |  | On Market (2024)FDACENMPA | unknown | unknown | unknown | PZT | Yes | ||
Delft Imaging | Low-cost CE-certified probe connected to a smartphone with solar-based smartphone charging |  | On Market (2021)CE | unknown | unknown | unknown | Unknown | Yes | ||
Clarius | Portable wireless U/S device with convex (curved) array specialized for fast imaging, together with the AI-powered app to deliver best imaging and cloud storage |  | On MarketFDACE | unknown | unknown | unknown | PZT | Yes | ||
Healson | USB type C portable ultrasound with low power consumption |  | On MarketNPMA CE; 2015 | unknown | unknown | unknown | PZT | No | ||
Sonostar | Wireless double-headed ultrasound probe with convex, linear, and phased array transducers |  | On MarketFDACENMPA | unknown | unknown | unknown | PZT | No | ||
Vinno | 3 separate probes with convex, linear, and phased array transducers. High-throughput hardware facilitates obtaining high-fidelity data in real-time without loss. |  | On MarketCE (2020)FDANMPA (2020) | unknown | unknown | unknown | Unknown | No | ||
Mindray | Wireless handheld ultrasound that can be connected to a smartphone | .png) .png) | On MarketFDA (2024) | unknown | unknown | unknown | PZT | No | ||
Pulsenmore | Home ultrasound attached to smartphone for live clinician guided and app-guided mode for tele-clinician review; single patient use to prevent cross-contamination |  | On MarketEU Israel Brazil; Investigational Use: US | unknown | unknown | unknown | Unknown | No | ||
Exo | 3-in-1 imaging (curved, phased, and linear arrays) in one probe |  | Under Development | unknown | unknown | unknown | PMUT | Yes | ||
Bloom Standard | The Bloom Standard device enables autonomous, hands-free ultrasound scans to be performed anywhere, by any user |  | Under Development | unknown | unknown | unknown | 3D CMUT | Yes | ||
Butterfly | A single-probe, whole-body point-of-care ultrasound solution with the patented Ultrasound-on-Chip™ technology. Fully integrated device, mobile app, AI and cloud for diagnosis. Offered ongoing software subscription of education, AI tools, diagnostic tools |  | On MarketFDA | unknown | unknown | unknown | CMUT | Yes | ||
GE Healthcare | A wireless handheld ultrasound in a pocket-sized, dual-probe solution (linear and curved arrays). Digital tools available through optional subscription: share scan screen with remote party, cloud based secure data storage for exams and images, device management system.) |  | On Market (2009)FDA | unknown | unknown | unknown | PZT | Yes | ||
Philips | Curved transducer optimized for all |  | On Market (2018)FDA | unknown | unknown | unknown | PZT | Yes |
Supporting Innovations
POCUS devices with integrated AI-based software are usable by non-medical personnel without extensive training, overcoming a common roadblock to health care in LMICs.
AI-Based Software
AI models enable non-medical care workers to obtain key obstetric information through “blind sweeps” with the ultrasound probe. These sweeps are unguided, require less than a day of training, do not need real-time feedback and a traditional image display, and they yield data for AI-supported automated analysis without the need for expert medical interpretation. This automated analysis leverages high-quality images collected by the hardware probe to output key fetal aspects, such as fetal number, fetal presentation, and gestational age.
Some portable ultrasound hardware companies are creating algorithms for obstetric indications. Other companies have algorithms for other ultrasound use cases (e.g., lung, cardiac). These companies could develop algorithms for obstetric indications. Finally, there are software companies that do not have dedicated ultrasound hardware. Instead, they are an AI vendor for other ultrasound hardware companies (e.g., Rayshape). Regulatory approval for these algorithms is an indication of their commercial potential, and a low price enables use in LMICs. These models can be assessed according to how representative the training dataset is to the actual population (e.g., image volume, patient volume).
The table provides an overview of the state of innovation, with the listed technologies serving as representative samples of products available in the market or in development. Expand full table view
Name | Innovator | Description | Image | Dev Status | Use Case | Regulatory Approval | Price | Model Training Dataset | Learn More |
|---|---|---|---|---|---|---|---|---|---|
Delft Imaging | Mobile application installed on the smartphone that uses 6 blind sweeps and scans acquired by any frontline worker in 2 minutes |  | On Market (LMICs) | OB - Getational age, fetal presentation, multple gestation, fetal heart rate, placenta location | CE mark | unknown | Unknown | ||
Clarius | Automatically highlights fetal anatomy and provides biometric measurements to estimate gestational age, weight, and growth intervals |  | On Market | OB (FDA cleared); only available in USA and Canada; Under Development: Cardiac, Lung | FDA - OB (2024), MSK (2023), Bladder (2023) | $595/year | 30k fetal ultrasound images | ||
Edan (in partnership with GH Labs) | Leverage Nano C5 hardware with novel software that restricts image visibility. User sees instructions for data collection and outputs. |  | Under Development (2027) | OB - Gestational age, fetal presentation, estimated fetal weight, multiple gestation, amniotic fluid volume | n/a | unknown | ~106k videos | ||
Intelligent Ultrasound (sold to GE) | Software for gestational age with ability to pair with a variety of ultrasound devices |  | Under Development | OB - Gestational age | Not licensed for clinical use (2025) | unknown | Unknown | ||
Exo | Compatible with Exo Iris device; Automated capture and AI-based indications for cardiac and lung. |  | On Market | Cardiac, Lung; Under Development: OB | FDA cleared - lung, cardiac, bladder | included in device price ($3.5k) | 100k images (cardiac & lung) | ||
Bloom Standard | 1-minute, autonomous ultrasound imaging scan for earlier detection, rapid assessment and monitoring of medical conditions in mothers, babies and children. Its initial suite of deep learning algorithms will help risk-stratify babies and children of risk of potentially life-threatening cardiac and cardiopulmonary conditions. |  | Under Development | OB, Lung, Cardiac | pending FDA approval | unknown | Unknown | ||
Philips | Building algorithms on blind sweep features and image acquisition to guide users | Under Development | OB | n/a | unknown | n/a | |||
Caption Health (acquired by GE Healthcare) | Caption AI: AI for real-time guidance and feedback on image quality to capture diagnostic quality images. Caption Interpretation: selects the best image to calculate ejection fraction for cardiac function. |  | On Market | Cardiac, Image Quality | FDA - image guidance, cardiac | $500/year | 4M images from 9k patients | ||
Rayshape | Experienced in integrating AI in U/S devices for companies like Philips, GE Health, Edan, Vinno, Hisense requiring 1-3 months for AI integration into device for clinical validation. Authorized 25k + U/S machines and 120k+ AI licenses |  | On Market | Integrated 30 models | unknown | 1M+ images |