{"title":"Feasibility and Effectiveness of a Low-Code AI Platform for Developing a Neonatal Multimodal Pain Classification Model.","authors":"Nannan Yang, Xiaosong Jiang, Xue Jin, Xinran Dai, Yuanjing Gu, Huiping Jiang, Liping Pu, Tingqi Shi","doi":"10.2147/JMDH.S531709","DOIUrl":null,"url":null,"abstract":"<p><strong>Background: </strong>Artificial intelligence (AI) has advanced neonatal pain recognition, yet a significant gap persists in translating complex algorithms into practical clinical applications. Low-code AI development platforms, which simplify and automate model creation, offer a potential solution to bridge this gap between research and bedside practice.</p><p><strong>Objective: </strong>This study aimed to explore the feasibility of constructing and validating a neonatal multimodal pain classification model using a commercial low-code AI development platform (EasyDL). The objective was to develop an accessible, cost-effective, and efficient method that empowers clinical professionals to create their own AI tools without extensive programming expertise.</p><p><strong>Methods: </strong>We uploaded 426 neonatal acute pain multimodal data segments to the EasyDL platform and trained a video classification model using its AutoML capabilities. The model underwent internal testing on a held-out dataset portion, followed by external validation on an independent prospective cohort. For external validation, we compared model performance against the N-PASS (Neonatal Pain, Agitation, and Sedation Scale) scores assessed by a senior nurse as the clinical gold standard.</p><p><strong>Results: </strong>The neonatal multimodal pain classification model developed on the platform showed strong performance. Internal validation achieved 89.6% accuracy and an 85.8% F1 score. External validation on unseen data reached 87.7% accuracy, with AUC exceeding 0.95 across all pain categories (no pain, mild pain, severe pain). The streamlined development process enabled seamless API deployment to an Android mobile device for clinical use.</p><p><strong>Conclusion: </strong>Developing a neonatal multimodal pain classification model using a low-code AI platform proves both feasible and effective. The model demonstrates robust performance and strong clinical integration potential. This approach offers a practical pathway to democratize AI development, enabling healthcare professionals to create digital solutions for neonatal pain management.</p>","PeriodicalId":16357,"journal":{"name":"Journal of Multidisciplinary Healthcare","volume":"18 ","pages":"5771-5780"},"PeriodicalIF":2.4000,"publicationDate":"2025-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12442815/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Multidisciplinary Healthcare","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.2147/JMDH.S531709","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2025/1/1 0:00:00","PubModel":"eCollection","JCR":"Q2","JCRName":"HEALTH CARE SCIENCES & SERVICES","Score":null,"Total":0}
引用次数: 0
Abstract
Background: Artificial intelligence (AI) has advanced neonatal pain recognition, yet a significant gap persists in translating complex algorithms into practical clinical applications. Low-code AI development platforms, which simplify and automate model creation, offer a potential solution to bridge this gap between research and bedside practice.
Objective: This study aimed to explore the feasibility of constructing and validating a neonatal multimodal pain classification model using a commercial low-code AI development platform (EasyDL). The objective was to develop an accessible, cost-effective, and efficient method that empowers clinical professionals to create their own AI tools without extensive programming expertise.
Methods: We uploaded 426 neonatal acute pain multimodal data segments to the EasyDL platform and trained a video classification model using its AutoML capabilities. The model underwent internal testing on a held-out dataset portion, followed by external validation on an independent prospective cohort. For external validation, we compared model performance against the N-PASS (Neonatal Pain, Agitation, and Sedation Scale) scores assessed by a senior nurse as the clinical gold standard.
Results: The neonatal multimodal pain classification model developed on the platform showed strong performance. Internal validation achieved 89.6% accuracy and an 85.8% F1 score. External validation on unseen data reached 87.7% accuracy, with AUC exceeding 0.95 across all pain categories (no pain, mild pain, severe pain). The streamlined development process enabled seamless API deployment to an Android mobile device for clinical use.
Conclusion: Developing a neonatal multimodal pain classification model using a low-code AI platform proves both feasible and effective. The model demonstrates robust performance and strong clinical integration potential. This approach offers a practical pathway to democratize AI development, enabling healthcare professionals to create digital solutions for neonatal pain management.
期刊介绍:
The Journal of Multidisciplinary Healthcare (JMDH) aims to represent and publish research in healthcare areas delivered by practitioners of different disciplines. This includes studies and reviews conducted by multidisciplinary teams as well as research which evaluates or reports the results or conduct of such teams or healthcare processes in general. The journal covers a very wide range of areas and we welcome submissions from practitioners at all levels and from all over the world. Good healthcare is not bounded by person, place or time and the journal aims to reflect this. The JMDH is published as an open-access journal to allow this wide range of practical, patient relevant research to be immediately available to practitioners who can access and use it immediately upon publication.