A deep transfer learning–based artificial intelligence model, considered one of the most advanced forms of machine learning, achieved 92% accuracy in predicting spoken language outcomes one to three years after cochlear implantation, according to a large international study published in JAMA Otolaryngology–Head & Neck Surgery.
Cochlear implants are currently the only effective intervention for children with severe to profound hearing loss to gain access to sound and spoken language. However, despite early implantation, language development outcomes vary widely when compared with children born with normal hearing. Early identification of children at risk for poorer language outcomes could allow clinicians to provide intensified therapy at an earlier stage.
To address this challenge, researchers trained AI models using pre-implantation brain MRI scans from 278 children across Hong Kong, Australia, and the United States. The children spoke English, Spanish, and Cantonese, and the participating centers employed different imaging protocols and outcome assessment methods.
While such heterogeneous datasets pose significant challenges for traditional machine learning approaches, the deep learning model demonstrated excellent performance and consistently outperformed conventional machine learning methods across all outcome measures.
“Our findings demonstrate the feasibility of using a single AI model as a reliable prognostic tool for predicting language outcomes in children receiving cochlear implants worldwide,” said senior author Nancy M. Young, MD, Medical Director of Audiology and Cochlear Implant Programs at Ann & Robert H. Lurie Children’s Hospital of Chicago.
The AI-driven approach enables a “predict-to-prescribe” strategy, allowing clinicians to tailor therapy intensity to optimize language development in individual children.
