METHODS: Two blinded reviewers searched PubMed, Embase, Scopus, Web of Science, and IEEE Xplore databases, then selected and graded the risk of bias of observational studies of adults (≥ 18 years) comparing the diagnostic performance of AI algorithms using craniofacial photographs, versus conventional OSA diagnostic criteria (i.e. apnea-hypopnea index [AHI]). Studies were excluded if they detected apneic events without diagnosing OSA. AI models evaluated with a random split test set or k-fold cross-validation were included in a Bayesian bivariate meta-analysis.

RESULTS: From 5,147 records, 6 studies were included, containing 10 AI models trained/tested on 1,417/983 participants. The risk of bias was low. AI trained on craniofacial photographs achieved a pooled 84.9% sensitivity (95% credible interval [95% CrI]: 77.1-90.7%) and 71.2% specificity (95% CrI: 60.7-81.4%). Bayesian meta-regression identified deep learning (convolutional neural networks) as the most accurate AI algorithm (91.1% sensitivity, 79.2% specificity) comparable to home sleep apnea tests. AHI cutoffs, OSA prevalence, feature engineering, input data, camera type and informativeness of Bayesian prior did not alter diagnostic accuracy. There was no substantial publication bias.

METHODS: Frontal view intraoral photographs fulfilling selection criteria were collected. Along the gingival margin, the gingival conditions of individual sites were labelled as healthy, diseased, or questionable. Photographs were randomly assigned as training or validation datasets. Training datasets were input into a novel artificial intelligence system and its accuracy in detection of gingivitis including sensitivity, specificity, and mean intersection-over-union were analysed using validation dataset. The accuracy was reported according to STARD-2015 statement.

RESULTS: A total of 567 intraoral photographs were collected and labelled, of which 80% were used for training and 20% for validation. Regarding training datasets, there were total 113,745,208 pixels with 9,270,413; 5,711,027; and 4,596,612 pixels were labelled as healthy, diseased, and questionable respectively. Regarding validation datasets, there were 28,319,607 pixels with 1,732,031; 1,866,104; and 1,116,493 pixels were labelled as healthy, diseased, and questionable, respectively. AI correctly predicted 1,114,623 healthy and 1,183,718 diseased pixels with sensitivity of 0.92 and specificity of 0.94. The mean intersection-over-union of the system was 0.60 and above the commonly accepted threshold of 0.50.

METHOD: In the present study, three separate data cohorts containing 1288 breast lesions from three countries (Malaysia, Iran, and Turkey) were utilized for MLmodel development and external validation. The model was trained on ultrasound images of 725 breast lesions, and validation was done separately on the remaining data. An expert radiologist and a radiology resident classified the lesions based on the BI-RADS lexicon. Thirteen morphometric features were selected from a contour of the lesion and underwent a three-step feature selection process. Five features were chosen to be fed into the model separately and combined with the imaging signs mentioned in the BI-RADS reference guide. A support vector classifier was trained and optimized.

RESULTS: The diagnostic profile of the model with various input data was compared to the expert radiologist and radiology resident. The agreement of each approach with histopathologic specimens was also determined. Based on BI-RADS and morphometric features, the model achieved an area under the receiver operating characteristic (ROC) curve (AUC) of 0.885, which is higher than the expert radiologist and radiology resident performances with AUC of 0.814 and 0.632, respectively in all cohorts. DeLong's test also showed that the AUC of the ML protocol was significantly different from that of the expert radiologist (ΔAUCs = 0.071, 95%CI: (0.056, 0.086), P = 0.005).

MATERIALS AND METHODS: An electronic search was carried out using databases such as PubMed, Scopus, and the Web of Science Core Collection. Two reviewers searched the databases separately and concurrently. The initial search was conducted on 6 July 2021. The publishing period was unrestricted; however, the search was limited to articles involving human participants and published in English. Combinations of Medical Subject Headings (MeSH) phrases and free text terms were used as search keywords in each database. The following data was taken from the methods and results sections of the selected papers: The amount of AI training datasets utilized to train the intelligent system, as well as their conditional properties; Unilateral CLP, Bilateral CLP, Unilateral Cleft lip and alveolus, Unilateral cleft lip, Hypernasality, Dental characteristics, and sagittal jaw relationship in children with CLP are among the problems studied.

RESULTS: Based on the predefined search strings with accompanying database keywords, a total of 44 articles were found in Scopus, PubMed, and Web of Science search results. After reading the full articles, 12 papers were included for systematic analysis.