Table 6.

Summary of included studies on surgical workflow and data analysis.

Study			Year		Data modality		Data source		Sample size		MLa method	Data availability
Learning curve analysis
	[101]	2024		Video		Simulator		N/Ab		3D CNNc; TSNd		Private
	[102]	2022		Kinematics; force		VRe simulator (NeuroVR)		50 participants		KNNf (for metric selection); ANOVA		Private
Video retrieval/content understanding
	[26]	2024		Image; text (VQAg)		EndoVis17/18; DAISIh-VQA		545 QAi pairs (DAISI)		Multimodal LLMj (InstructBLIP); CLk		Public
	[103]	2024		Video		Cholec80l; Cataract-101; CholecT45m		80, 101, 45 videos		Hashing; Transformer		Public
	[25]	2022		Video		Surgical Actions 160; Cataract-101		160 clips; 101 surgeries		CNN; LSTMn; TCNo; contrastive learning		Public
	[104]	2019		Kinematics		JIGSAWSp		8 participants		DTWq; DBAr; NLTSs		Public

^aML: machine learning.

^bN/A: not available.

^cCNN: convolutional neural network.

^dTSN: temporal segment network.

^eVR: virtual reality.

^fKNN: k-nearest neighbors.

^gVQA: visual question answering.

^hDAISI: Database for AI Surgical Instruction.

ⁱQA: question-answer.

^jLLM: large language model.

^kCL: contrastive learning.

^lCholec80: a dataset of 80 cholecystectomy procedure videos.

^mCholecT45: a dataset of 45 cholecystectomy procedure videos with triplet annotations.

ⁿLSTM: long short-term memory.

^oTCN: temporal convolutional network.

^pJIGSAWS: Johns Hopkins University and Intuitive Surgical Inc Gesture and Skill Assessment Working Set.

^qDTW: dynamic time warping.

^rDBA: DTW barycenter averaging.

^sNLTS: nonlinear time series.