Back to 2026 Abstracts
Using Artificial Intelligence To Analyze Speech Outcomes In Patients With Cleft Lip And Palate: A Pilot Program In Columbia
Rachel Donaldson, BS, MS1, Priyanka Naidu, MD
2, Mark Urata, MD
1, William Magee, MD
3, Caroline Yao, MD
2.
1Children's Hospital of Los Angeles, Los Angeles, CA, USA,
2Division of Plastic and Reconstructive Surgery, Keck School of Medicine of USC, Los Angeles, CA, USA,
3Shriners Children's Hospital of Southern California, Los Angeles, CA, USA.
PURPOSE: In LMICs, limited access to speech-language pathologists and delayed diagnosis of velopharyngeal insufficiency (VPI) contribute to language delays, communication barriers, and stigma. Artificial intelligence (AI) offers a potential solution by analyzing speech recordings from a smartphone app, trained on cleft-focused SLP evaluations, to detect errors and guide triage for therapy or surgical referral. This pilot program explores feasibility of an AI-enabled tool in Colombia with Operation Smile.
METHODS: Speech recordings and palatal photographs were collected from children ≥ 4 with repaired or unrepaired CLP and from controls. Using a smartphone app, participants repeated 16 standardized words targeting key phonemes. Demographics and cleft characteristics were recorded. Five cleft-trained SLPs independently evaluated each sample using ASHA-based criteria to grade VPI and identify articulation errors. Based on these evaluations, SLPs recommended triage needs, including speech therapy, surgical referral, or both. Annotated recordings with SLP assessments were then provided to an AI team to train an algorithm to detect VPI, grade severity, and guide triage.
RESULTS: To date, 26 participants aged 4-18 with repaired or unrepaired CLP and one control have been enrolled. Standardized speech recordings, palatal photographs, and demographic data were collected, and cleft-focused SLPs are independently reviewing recordings to create annotated datasets for algorithm training.
CONCLUSION: This project demonstrates the feasibility of using a smartphone app to collect standardized speech samples for AI-based evaluation in a low-resource setting. Ongoing work will support development of an AI algorithm to detect VPI, grade severity, and guide triage for therapy or surgical intervention.
Back to 2026 Abstracts
