Columbia Technology Ventures

Ultrasound based, deep-learning algorithm for preterm birth risk prediction

This technology is a deep learning algorithm that uses ultrasound imaging data to extract cervix geometry and predict preterm birth (PTB) risk.

Unmet Need: Accurate, reliable prediction of preterm birth risk

The current gold standard for prediction of preterm birth (PTB) risk relies on clinical assessment of transvaginal ultrasound (TVUS) images to measure cervical length (CL). However, this method is an incomplete measurement of cervical health and is inconsistent, especially for patients with no prior history of PTB or who are pregnant for the first time. There is currently no automated, objective, and quantitative method to accurately assess PTB risk.

The Technology: Deep-learning algorithm to accurately predict preterm birth from transvaginal ultrasound images

This technology uses machine learning to automatically detect and label cervical physiological features from TVUS images of patients in the 2nd and 3rd trimesters of pregnancy, records PTB markers and other important cervical and lower uterine features, combines additional patient data from electronic medical records with TVUS sonograms, and applies deep learning models to predict PTB risk. The algorithm has been trained and tested on a data set from Columbia patients and successfully recognized cervical features across multiple cervical phenotypes. As such, this technology may be used to improve prediction of PTB.

Applications:

  • Clinical tool for PTB prediction
  • Diagnostic tool for cervical health
  • Diagnostic tool for overall vagina health
  • Tracking tool for pregnancy management
  • Platform for OB/GYN research
  • Education tool for TVUS sonogram practitioners

Advantages:

  • Automated and unbiased assessment of TVUS images
  • Complete assessment of cervical health
  • Potential for higher diagnostic accuracy
  • Fits into existing clinical workflow

Lead Inventor:

Kristin M. Myers, Ph.D.

Patent Information:

Patent pending

Related Publications:

Tech Ventures Reference: