Table 8.

Limitations of relevant studies employing AI for disease diagnosis.

Diagnostic Task	Target Problem	Author, Year (Ref.)	Limitations
Disease diagnosis	bone recession and interradicular radiolucency	Khan et al., 2021 [44]	Limited in terms of size
	temporomandibular joint disorders	Orhan et al., 2021 [36]	Further study is required for identification for the reduction state.
		Choi et al., 2021 [58]	The dataset size is rather small and more images are required to improve performance
	maxillary sinusitis lesions	Hung et al., 2022 [55]	The performance can be further improved
		Kuwana et al., 2022 [56]	Dataset is small in terms of size. The study also did not include post-operative maxillary sinuses
		Kim et al., 2019 [42]	Dataset is limited in terms of size and only includes maxillary sinus using water view radiographs
		Murata et al., 2019 [57]	Limited dataset in terms of size
	alveolar bone delineation	Duong et al., 2019 [45]	Limited dataset in terms of size
	bone assessment	Nguyen et al., 2020 [46]	The model is restricted to detection and segmentation in buccal surfaces
	periapical lesions	Orhan et al., 2020 [48]	The model can be influenced with variations i.e., presence of endo-perio lesions or other periodontal defects
	bone loss	Lee et al., 2022 [41]	The model is not able to detect vertical defect depth and angulation. The diagnosis also relies on the reliability of examiners
		Kim et al., 2019 [63]	Low resolution for the individual tooth in panoramic images as it captures a wide field of view
		Krois et al., 2019 [5]	Limited dataset in terms of size
		Zheng et al., 2021 [8]	The approach is not applicable on unlabeled data
		Papantonopoulos et al., 2014 [23]	The model is not flexible enough to capture nonlinearities in data
	oral cancer	Jeyaraj et al., 2019 [59]	Dataset is limited in terms of size
	odontogenic cystic lesion	Lee et al., 2019 [62]	Limited dataset in terms of size
	periodontal bone destruction	Moran et al., 2020 [7]	Poor performance in classifying healthy regions due to small dataset
	vertical root fracture	Fukuda et al., 2020 [49]	The model is trained on panoramic radiographs with clear vertical root fracture (VRF) lines which impacts the performance
	apical lesions	Ekert et al., 2019 [50]	Manually cropped image segments were used for training The sensitivity should be improved before clinical use
	interproximal caries	Bayraktar & Ayan, 2022 [51]	Dataset is limited in terms of size. Carious lesions were not classifiedas enamel caries or dental caries
	distal root assessment	Hiraiwa et al., 2019 [52]	Image patches are created by manual segmentation which is time-consuming.
	proximal and occlusal caries	Casalegno et al., 2019 [53]	Limited dataset in terms of size and ground truth labels
	dental implants	Lee et al., 2020 [43]	The proposed study only includes three types of dental implants which limit its practical use
	preserve tooth boundary	Takahashi et al., 2021 [71]	Limited dataset in terms of size of maxilla set
Disease diagnosis	tooth decay detection	Geetha et al., 2018 [19]	Small dataset The model does not provide classification based on caries depth
	alveolar bone loss detection	Lin et al. 2015 [6]	Poor performance on unevenly illuminated images
	occlusal caries lesion detection	Berdouses et al., 2015 [32]	The accuracy can be improved further
	sagittal skeletal patterns identification	Nino-Sandoval et al., 2016 [34]	Limited class discrimination
	vertical root fracture identification	Kositbowornchai et al., 2013 [20]	The proposed model is not applicable on other root fractures in clinical practice
	scoring lesions	Ghaedi et al. 2014 [31]	Small dataset Unbalanced number of images, histological verification of the diagnosis was not performed
	cervical vertebra maturation degree	Budiman, 2013 [22]	There is an overlap in data distribution that affects the models’ performance
	facial attractiveness	Yu et al., 2014 [35]	The model prediction is limited to certain angles and ratios
	plaque segmentation	Li et al., 2022 [29]	The model is not generalizable to other caries areas
	detect vertical vertebra maturation degree	Amasya et al. 2020 [21]	Hand-wrist radiographs are not considered
	maxillary structure assessment	Chen et al., 2020 [33]	The dataset is limited in terms of size