Role of Artificial Intelligence in Cleft Lip and/or Cleft Palate in Diagnosis and Detection—An Umbrella Review
Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine
ABSTRACT Purpose Early diagnosis and treatment are essential in managing congenital cleft conditions. Use of artificial intelligence (AI) in routine diagnosis can substantially improve management of chronic conditions; however, in the context of CL/P, evidence remains fragmented. Thus, an umbrella review was planned to explore the applications of AI‐based diagnostic systems in managing orofacial clefts. cre270386-sec-0010 Methods Priori protocol was registered with PROSPERO. Five electronic databases were thoroughly searched (PubMed/MEDLINE, Scopus, Embase, Google Scholar, ScienceDirect), based on the pre‐defined PICO framework. The data extraction form was designed in accordance with the Joanna Briggs Institute (JBI) guidelines and analyzed. Results were presented in the form of tables, supported by narrative summaries. Overlap assessment was conducted to avoid overemphasis and duplication of the overall results. Methodological robustness of the included studies was assessed using AMSTAR 2.0 tool. cre270386-sec-0020 Results Of 395 initially retrieved articles, only three systematic reviews were included for this study. Overlap assessment indicated a high percentage of overlap among studies, with the corrected covered area to be 13.89%. Overall analysis revealed that AI models—DCNN (97% to 96%), RF (99% to 96%), and SVM (94% to 93%) demonstrated high accuracy in diagnosing orofacial clefts. Deep learning models were extensively used in diagnosing and predicting the orofacial
Abstract
ABSTRACT Purpose Early diagnosis and treatment are essential in managing congenital cleft conditions. Use of artificial intelligence (AI) in routine diagnosis can substantially improve management of chronic conditions; however, in the context of CL/P, evidence remains fragmented. Thus, an umbrella review was planned to explore the applications of AI‐based diagnostic systems in managing orofacial clefts. cre270386-sec-0010 Methods Priori protocol was registered with PROSPERO. Five electronic databases were thoroughly searched (PubMed/MEDLINE, Scopus, Embase, Google Scholar, ScienceDirect), based on the pre‐defined PICO framework. The data extraction form was designed in accordance with the Joanna Briggs Institute (JBI) guidelines and analyzed. Results were presented in the form of tables, supported by narrative summaries. Overlap assessment was conducted to avoid overemphasis and duplication of the overall results. Methodological robustness of the included studies was assessed using AMSTAR 2.0 tool. cre270386-sec-0020 Results Of 395 initially retrieved articles, only three systematic reviews were included for this study. Overlap assessment indicated a high percentage of overlap among studies, with the corrected covered area to be 13.89%. Overall analysis revealed that AI models—DCNN (97% to 96%), RF (99% to 96%), and SVM (94% to 93%) demonstrated high accuracy in diagnosing orofacial clefts. Deep learning models were extensively used in diagnosing and predicting the orofacial cleft with accuracies across models ranging over 90%. Machine learning models also demonstrated good performance in identifying genetic risk. However, the lowest accuracies were demonstrated by the DesNet model (nearly 73% accuracy). Also, methodological robustness indicated a moderate level of confidence, suggesting limitations in the generalizability of the findings. cre270386-sec-0030 Conclusions Findings present the potential of AI models in diagnosing and detecting orofacial clefts. AI models can support the early diagnosis of orofacial clefts; however, this study highlights the need for more comprehensive research to determine how different AI models can improve early diagnosis and treatment outcomes. CRD420251125934. cre270386-sec-0040