Table 3.

The application of AI in image-guided techniques

Study	Aim	AI method	N	Acuraccy results	Conclusions
Alkhatib 2019 [42]	Median and sciatic nerve tracking in ultrasound images.	CNN	42	Accuracy, 0.87	Models showed superior performance and handled noise suppression without pre-filtering the images
Hetherington 2017 [17]	Identify lumbar vertebral levels in US images and display it with augmented reality.	CNN	20	Accuracy, 85%	The system successfully identifies lumbar vertebral levels.
InChan 2021 [18]	Determine the needle insertion point in obese patients using US image.	Machine-learning	48	Success rate for spinal anesthesia on first attempt 79.1%	The program is able to provide assistance to needle insertion point identification in obese patients.
Liu 2021 [20]	Locate, from US images, the anesthesia point of patients with regional nerve block.	CNN	100	Higher positioning accuracy and lower postoperative complications	The model can effectively improve the accuracy of US images.
Pesteie 2018 [43]	Automatically localize the needle target for epidural needle placement in US of the spine.	CNN	20	Average lateral and vertical error 1 mm, 0.4 mm, respectively.	The algorithm average errors are inferior to the clinically acceptable error.
Yoo 2021 [21]	Interpret video bronchoscopy images of the carina and main bronchi	CNN	180	Accuracy, 0.84; AUC 0.9752	This model can be a basis for designing a clinical decision support system with video bronchoscopy
Yu 2015 [44]	Identify the bone/interspinous region for US images obtained from pregnant patients.	SVM	20	Accuracy, 93.2 PPV, 94.17 Sensitivity, 93.05 AUC 97.55	Proposed method can process the ultrasound images of lumbar spine in an automatic manner.
Yusong 2016 [19]	Determine the needle entry site for epidural anesthesia in real time	SVM	53	Accuracy 0.94	Even the anesthetists with little experience in US could determine the suitable puncture site accurately and efficiently.

US ultrasonography, CNN convolutional neural network, SVM support vector machine