Cribsy Phase I: Successful pilot sets up clinical validation

Between mid-2024 and 2025, with support from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), Early Markers ran the first funded test of Cribsy: could parents at home, using nothing but a phone, produce the video an AI system needs to screen an infant for neuromotor risk, and could we take that screen to the communities that need it most? Here is what the Phase I project delivered.

The platform

• A working app, in English and Spanish. Cribsy's beta was released as a web app and as native iOS and Android apps. Parents receive prompts, record short videos of their infant in set postures, and upload them from home.
• AI that reads infant movement. We built and tuned models for infant pose estimation and for classifying neuromotor risk from the resulting movement features, and an analysis pipeline that processes uploaded videos automatically. By the end of Phase I we had automated the retraining loop so the models can keep improving as data accrues.
• A standardized assessment, automated. To show the platform extends beyond a single screen, we automated the Alberta Infant Motor Scale, a widely used instrument for infants from 3 to 18 months.
• Explainability. Clinicians need to know why a screen flagged a child. We demonstrated explainable-AI reports that show which movement features drove a classification, and added a chatbot built from CDC developmental guidance to answer parents' questions inside the app.

The clinics

The part of Phase I we are proudest of happened away from the keyboard. To reach infants in rural and Native communities, we ran four pop-up screening clinics in Washington and Kansas, including two on tribal lands with the Lummi Nation and the Nooksack Tribe. Each clinic was organized with a local Patient Navigator, a trusted community member who recruited families, arranged the venue and translation, helped parents record and upload videos, and followed up afterward with families whose infants were referred for further evaluation. Forty infants took part. Every participating parent was able to record and upload multiple home videos when prompted, which answered the question we most needed answered.

Cribsy's performance numbers

On the Phase I data, a compact set of twenty movement features, kinematic measures plus the infant's age, distinguished at-risk from typically developing infants with a sensitivity of 92% and a specificity of 77.5%. Those are pilot figures from a small sample and they exist to design the next study, not to make clinical claims. Their practical effect was to shrink the number of infants the Phase II validation study needs, which is good news for the families and clinics involved.

Toward the pediatrician's office

Phase I also looked hard at how Cribsy fits real pediatric practice: workflow, reimbursement, and regulatory pathways. Multiple screens used together improve accuracy and reduce unnecessary referrals, so we positioned Cribsy as an objective second screen alongside questionnaire-based tools, and began a collaboration with CHADIS, a platform used in over 250 pediatric practices that collects hundreds of thousands of parent questionnaires each month.

What comes next

Phase I ended with a patent issued on Cribsy's core method and a Phase II application that reviewers scored well. That application has now been funded, and the next post describes what the next three years look like.

Research reported here was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health under Award Number R43HD115478. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.