Table 3.
Summary of included studies on skill classification and regression.
| Study | Year | Data modality | Data source | Sample size | Assessment target/outcome | MLa method | Data availability |
| [33] | 2025 | Video | Simulator | N/Ab | Laparoscopic skill validation | YOLOv4c | Private |
| [34] | 2025 | Video | Simulator | N/A | Technical skill classification | Vision Transformer | Private |
| [35] | 2025 | Text (free-text comments) | Simulator (Vertex Pursuit) | 600 data points (24 participants) | Skill level classification (novice, intermediate, and expert) | LLMd (Claude 3 Opus, GPT-4, Gemini Pro); BERTe; SVMf; KNNg; RFh; XGBoosti; MLPj | Private |
| [36] | 2025 | Video | Clinical (colorectal) | N/A | Skill assessment | CNNk (EfficientNetB7) | Private |
| [37] | 2025 | Motion (rotation) | Simulator | N/A | Suturing skill assessment (hand rotation) | Functional data analysis | Private |
| [38] | 2025 | Video (tool motion) | Simulator (EA OJTl Model) | 45 videos | Skill level classification (good vs poor) | YOLOv3; CNN (4-layer) | Private |
| [39] | 2025 | Video | Clinical (cataract) | >1000 videos | OSACSSm performance ratings | YOLACTn | Private |
| [10] | 2024 | Kinematics; force | VRo/ARp simulator | 27 participants | Expertise classification | ANNq; transfer learning | Private |
| [40] | 2024 | Video (surgestures) | Clinical (laparoscopic cholecystectomy) | 75 videos (33 surgeons) | mGOALSr skill classification (competent vs incompetent) | SVM; LRs; RF; GBDTt; AdaBoost | Private |
| [41] | 2024 | Video | EndoVis SkillNet | N/A | Skill classification | Video Transformer (multitask) | Public |
| [42] | 2024 | Video | Suture simulator | 314 videos | GRSu score classification (novice, intermediate, and proficient) | Video Swin Transformer; I3Dv | Public |
| [43] | 2024 | Video | Clinical (laparoscopic distal gastrectomy) | 256 videos (training); 180 videos (testing) | ESSQSw skill score (phase duration and AICSx) | CNN (EfficientNetB7) | Private |
| [44] | 2024 | Video | Clinical (laparoscopic colorectal) | 1572 videos | ESSQS skill score regression | EfficientNetB7; regression | Private |
| [24] | 2023 | Video (motion) | Laparoscopic trainer (LMICsy) | 74 videos (47 trainees) | OSATSz score prediction | KNN; YOLOv5 | Private |
| [45] | 2024 | EEGaa; eye-tracking | Clinical (porcine model) | 43 surgeries (11 surgeons) | GEARSab skill level classification | RF; XGBoost; LR; gradient boosting | Private |
| [46] | 2024 | Force | Simulator | N/A | Skill assessment (Smoothness of Force) | Time/frequency analysis | Private |
| [47] | 2024 | Motion (EMac sensors) | SutureCoach Simulator | 97 participants | Expertise level classification (novice, intermediate, and expert) | Statistical metrics | Private |
| [48] | 2024 | Kinematics | dVRKad | N/A | Bimanual coordination assessment | Information theory; DTWae | Private |
| [30] | 2024 | Video | Clinical (cataract surgery) | 197 videos (12 surgeons) | Skill level classification (expert vs novice) | 2D CNN-LSTMaf; 3D CNN | Private |
| [49] | 2024 | Kinematics | JIGSAWSag; Wolf dataset | 8 (JIGSAWS); 32 (Wolf) | GRS item score prediction (GOALSah/OSATS) | Multiheaded MLP | Public |
| [9] | 2024 | Kinematics (as images) | JIGSAWS | 8 participants | Skill level classification | Vision Transformer; CWTai | Public |
| [50] | 2023 | Video | JIGSAWS | 8 participants | GRS score prediction (relative to expert) | Transformer; TCNaj; ResNetak-18 | Public |
| [51] | 2023 | Video | Clinical (laparoscopic sigmoidectomy) | 120 videos | ESSQS score correlation (AICS) | CNN (Xception) | Private |
| [52] | 2023 | Kinematics | JIGSAWS | 8 participants | Skill level classification (novice, intermediate, and expert) | Dempster-Shafer theory; SVM; XGBoost | Public |
| [2] | 2023 | EEG | VR simulator (NeuroVR) | 21 participants | Expertise level classification (skilled vs less-skilled) | ANN; SVM; LDAal; RF; KNN; LR; naive Bayes | Private |
| [53] | 2023 | Video (tool motion) | JIGSAWS | 24 videos (8 participants) | Skill level classification (2-class and 3-class) | ResNet; KCFam tracking | Public |
| [54] | 2023 | Simulator metrics | ANGIO Mentor Simulator | 22 participants | Modified Reznick scale score prediction | Feedforward neural network | Private |
| [16] | 2023 | Kinematics | JIGSAWS; VR simulator | 8 (JIGSAWS); 16 (VR) | Skill level classification (novice vs expert) | CNN; Bi-LSTMan; MCao Dropout | Public |
| [55] | 2023 | Video (tool motion) | JIGSAWS; FLSap trainer | 8 (JIGSAWS); 33 (FLS) | Pass/fail classification | Res-CNNaq; DAEar; Mask R-CNNas; CAMat | Public |
| [56] | 2022 | Motion capture | Wet Lab box trainer | 89 sessions (70 participants) | GOALS score prediction | SVRau; PCAav; Ridge; PLSaw Regression | Private |
| [57] | 2022 | Video (hand motion) | Laparoscopic simulator | 9 participants | Hand movement performance assessment | Fuzzy logic; SSDax ResNet50 | Public |
| [58] | 2022 | Video (tool motion) | Cholec80 | 80 videos (13 surgeons) | GOALS efficiency score classification | Transformer; random forest; YOLOv5 | Public |
| [59] | 2022 | Kinematics; image | Knot-tying simulator | 360 trials (72 trainees) | OSATS score prediction (4 domains) | ResNet; Bi-LSTM | Public |
| [60] | 2022 | VR simulators | VR simulators | 254 sessions | GEARS score classification (novice, intermediate, and expert) | DNNay (Google AutoML) | Private |
| [27] | 2022 | fNIRSaz | FLS Trainer | 474 trials (7 students) | Pass/fail classification (FLS task) | Dilated causal CNN (WaveNet-style) | Private |
| [61] | 2022 | Motion; force | Phantom; clinical | 12 surgeons | Skill level classification (expert vs novice) | RF; XGBoost; SHAPba | Private |
| [62] | 2022 | Kinematics (1D) | Clinical (porcine model) | 100 trials (10 surgeons) | Intra- and intersubject skill similarity | Dynamic time warping | Private |
| [4] | 2021 | Kinematics; force | VR Simulator (Sim-Ortho) | 23 participants | Expertise classification | ANN (multilayer perceptron) | Private |
| [63] | 2021 | Kinematics | JIGSAWS | 33 trials (7 surgeons) | Skill level classification (novice, intermediate, and expert) | KNN; PCA; Relief | Public |
| [64] | 2021 | Video (optical flow) | JIGSAWS | 8 participants | Skill level classification (novice vs expert) | ResNet; CNN; LSTM; ConvAuto | Public |
| [65] | 2021 | Video (tool motion) | Clinical (cholecystectomy) | 949 clips | Likert scale skill score prediction (1-5) | Linear regression; ResNet50-FPNbb | Private |
| [66] | 2021 | Video; kinematics | JIGSAWS; Clinical (laparoscopic gastrectomy) | 206 videos (JIGSAWS); 20 videos (clinical) | GRS score prediction (modified OSATS) | TCN; ResNet-101; MS-TCNbc; MLP; contrastive learning | Public, Private |
| [67] | 2021 | Video | JIGSAWS | 8 participants | GRS score classification (expert vs novice) | SVM; BoFbd (STIPbe, iDTbf) | Public |
| [12] | 2021 | Video | JIGSAWS | 8 participants | Skill level classification (novice, intermediate, and expert) | ResNet50; FFTbg; 1D CNN | Public |
| [68] | 2021 | Sensor (force, time) | FRSbh Dome Simulator | 27 performances | FRS performance score (1-3) | ANFISbi; fuzzy logic | Private |
| [32] | 2020 | Kinematics | JIGSAWS | 8 participants | Skill level classification | CNN; LSTM; PCA; DFTbj; DCTbk; Autoencoder; SVM | Public |
| [69] | 2020 | Video | JIGSAWS | 8 participants | OSATS score prediction | I3D; TSNbl (multitask) | Public |
| [70] | 2020 | Video (tool motion) | Clinical (robotic thyroidectomy); Simulator (BABA Model) | 54 video segments | OSATS/GEARS skill classification (novice, skilled, and expert) | Mask R-CNN; Deep SORTbm; SVM; random forest | Private |
| [71] | 2020 | Kinematics | JIGSAWS | 8 participants | GRS score classification (expert vs novice) | Kernel SVM | Public |
| [72] | 2019 | Kinematics | JIGSAWS | 8 participants | OSATS score prediction | FCNbn; CAM | Public |
| [73] | 2019 | Video; optical flow | JIGSAWS | 8 participants | Skill level classification (novice, intermediate, and expert) | 3D CNN (I3D); TSN | Public |
| [74] | 2019 | Video (tool motion) | UVAbo Dataset | 12 surgeons | Skill level classification (novice vs expert) | HMMbp; SVM; LDA | Public |
| [75] | 2019 | Video (tool motion, optical flow) | Clinical (cataract surgery) | 99 videos | ICO-OSCARbq score prediction (expert vs nonexpert) | TCN | Private |
| [76] | 2019 | EMGbr; motion | Myo Armband (box trainer) | 99 knots (28 participants) | OSATS score prediction | Neural network; decision trees | Private |
| [77] | 2019 | Video (color, semantic) | Clinical (laparoscopic gastrectomy) | 57 videos (1 surgeon) | OSATS/COFbs score prediction | ResNet-101; regression; rank loss | Private |
aML: machine learning.
bN/A: not available.
cYOLOv4: You Only Look Once version 4.
dLLM: large language model.
eBERT: Bidirectional Encoder Representations from Transformers.
fSVM: support vector machine.
gKNN: k-nearest neighbors.
hRF: random forest.
iXGBoost: extreme gradient boosting.
jMLP: multilayer perceptron.
kCNN: convolutional neural network.
lEA OJT: esophageal atresia off-the-job training.
mOSACSS: Objective Structured Assessment of Cataract Surgical Skill.
nYOLACT: You Only Look at Coefficients.
oVR: virtual reality.
pAR: augmented reality.
qANN: artificial neural network.
rmGOALS: Modified Global Objective Assessment of Laparoscopic Skills.
sLR: logistic regression.
tGBDT: gradient boosted decision trees.
uGRS: Global Rating Scale.
vI3D: Inflated 3D ConvNet.
wESSQS: Endoscopic Surgical Skill Qualification System.
xAICS: Artificial Intelligence Confidence Score.
yLMIC: low-and-middle-income country.
zOSATS: Objective Structured Assessment of Technical Skills.
aaEEG: electroencephalography.
abGEARS: Global Evaluative Assessment of Robotic Skills.
acEM: electromagnetic.
addVRK: da Vinci Research Kit.
aeDTW: dynamic time warping.
afLSTM: long short-term memory.
agJIGSAWS: Johns Hopkins University and Intuitive Surgical Inc Gesture and Skill Assessment Working Set.
ahGOALS: Global Objective Assessment of Laparoscopic Skills.
aiCWT: continuous wavelet transform.
ajTCN: temporal convolutional network.
akResNet: residual network.
alLDA: linear discriminant analysis.
amANGIO: angiotherapy.
anBi-LSTM: bidirectional long short-term memory.
aoMC: Monte Carlo.
apFLS: Fundamentals of Laparoscopic Surgery.
aqRes-CNN: residual convolutional neural network.
arDAE: denoising autoencoder.
asMask R-CNN: mask region-based convolutional neural network.
atCAM: class activation mapping.
auSVR: support vector regression.
avPCA: principal component analysis.
awPLS: partial least squares.
axSSD: single shot detector.
ayDNN: deep neural network.
azfNIRS: functional near-infrared spectroscopy.
baSHAP: Shapley additive explanations.
bbFPN: feature pyramid network.
bcMS-TCN: multistage temporal convolutional networks.
bdBoF: bag of features.
beSTIP: spatio-temporal interest points.
bfiDT: improved dense trajectories.
bgFFT: fast Fourier transform.
bhFRS: Fundamentals of Robotic Surgery.
biANFIS: adaptive neuro-fuzzy inference system.
bjDFT: discrete Fourier transform.
bkDCT: discrete cosine transform.
blTSN: temporal segment network.
bmSORT: simple online and realtime tracking.
bnFCN: fully convolutional network.
boUVA: urethro-vesicle anastomosis.
bpHMM: hidden Markov model.
bqICO-OSCAR: International Council of Ophthalmology’s Ophthalmology Surgical Competency Assessment Rubric.
brEMG: electromyography.
bsCOF: clearness of operating field.