第23屆 使用AI系統來建立發展遲緩兒童早療效益的預測模式

使用AI系統來建立發展遲緩兒童早療效益的預測模式
Implementing a Real-time AI System to Predict the Progression of Gross Motor Developmental Delay in Children 1 Year Later

郭思辰¹、劉天慧²、張鈴艷²、劉忠峰^3*、周偉倪^2,4*、馬郁珊³、周琪¹、賴明琪¹
Szu-Chen Kuo¹, Tin-Wai Lau², Lin-Yen Chang², Chung-Feng Liu^3*, Willy Chou^2,4*
Yu-Shan Ma³, Julie Chi Chow¹, Ming-Chi Lai¹

¹奇美醫學中心兒科部、²奇美醫學中心復健部、
³奇美醫學中心醫學研究部門、⁴佳里奇美醫院復健科

¹ Department of Pediatric, Chi Mei Medical Center, Tainan, Taiwan; ² Department of Physical Medicine and Rehabilitation, Chi Mei Medical Center, Tainan, Taiwan;
³ Department of Medical Research, Chi Mei Medical Center, Tainan, Taiwan; ⁴ Department of Physical Medicine and Rehabilitation, Chi Mei Medical Center, Chiali, Tainan, Taiwan.

Background and Purpose:

Developmental delays may result in developmental disabilities which could greatly burden the families of affected children and the special education system. If they can receive early intervention earlier, they can have greater developmental gains. Previous studies have shown that motor development is a nonlinear process, and a long-term developmental tracking program is required to predict its trajectory. Currently, the available test scales can only present the current condition of the child and are unable to predict how the child will function in the future after treatment intervention. Because of this, it is very difficult for the healthcare team to develop a comprehensive and precise individualized early intervention program for children with developmental delay.

With the advancement of artificial intelligence (AI) and big data analysis technology, the case data accumulated from a large amount of experience, through machine learning methods could be used to develop new personalized and accurate forecasting tools, providing a long-term evaluation and potentially solving the existing problem.

The purpose of this study is to establish a personalized and accurate model for predicting the motor development milestone of children with developmental delay based on big data containing past developmental assessments, and to implement a convenient information system that can be used by healthcare staff in real-time. This research aims to formulate a predictive model for screening 1-year-later developmentally delayed children through artificial intelligence (AI)/machine learning (ML) to efficiently help healthcare workers in making clinical decisions.

Methods:

We identified 519 children who were evaluated and diagnosed with developmental delay from a single joint evaluation center of Children Joint Evaluation Centers of Chi Mei Medical Center in Taiwan from January 2016 to December 2019. The age of enrollment was from 3 months old to 6 years old. The evaluation test was Comprehensive Developmental Inventory for Infants and Toddlers (CDIIT).

Ten feature variables were used to construct an AI model to predict children with 1-year-after gross motor developmental delay. Five machine learning algorithms were used to establish this prediction model. The performance of the models was also evaluated in terms of accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). A prediction system was implemented according to the AI model with the highest AUC.

Results:

The Multilayer Perceptron (MLP) performed the best AUC for 1-year-after Comprehensive Developmental Inventory for Infants and Toddlers (CDIIT) Gross Motor Score prediction with an accuracy of 0.769, a sensitivity of 0.794, a specificity of 0.750, and an AUC of 0.868. We further implemented an AI prediction system with the best model to assist healthcare workers with their decision-making in real-time which received high acceptance after a simplified pilot survey.

Conclusion:

Machine learning prediction models provide a novel way to predict the possible developmental outcome of a child with developmental delay. With the prediction model, healthcare workers and parents of these children could obtain more objective information about child’s developmental condition including potential functional limitations. Appropriate interventions could be assessed and recommended in advance. Besides, we can adjust the variables for individual suggestions. We also use this system to establish an APP for care-givers’ education and enhancement of early intervention motivation. With the prediction system, therapists can discuss with the child’s parents for future interventions and potential functional limits. The prediction system is an ideal tool for shared decision-making between the healthcare provider and the patients and their family.

Keywords: Developmental delay, early intervention, artificial intelligence, gross motor outcome, prediction system