Table 1.
Evaluation metrics and clinical outcomes of artificial intelligence models in neurosurgery diagnosis and treatment.
| Author, Year, Country | Specialty | AI Model Types Used In the Study | Evaluation Metrics and Clinical Outcomes |
|---|---|---|---|
| Merali et al, 2021, Canada6 | Spinal Neurosurgery | DL (CNN) | Cervical Spinal Cord Compression Detection: Accuracy: 94% Sensitivity: 88% Specificity: 89% |
| Hallinan et al, 2022, Singapore7 | Spinal Neurosurgery | DL (CNN) | Spinal Metastases Detection: Internal test sets: Sensitivity: 97.6% Specificity: 93.6% External test sets: Sensitivity: 89.9% Specificity: 98.1% |
| Doerr et al, 2022, United States8 | Spinal Neurosurgery | DL (CNN) | Injury Classification Accuracy: 86.8% |
| Kim et al, 2020, Republic of South Korea9 | Spinal Neurosurgery | ML (Random forest, XGBoost, Bayesian generalized linear model, decision-making tree model, k-cluster analysis, logistic regression analysis and neural network analysis) | Operation time Accuracy: 97.5% Reoperation occurrence Accuracy: 95.2% |
| Hopkins et al, 2020, United States10 | Spinal Neurosurgery | ML (DNN) | Prediction of Postoperative SSI Accuracy: 78.7% |
| De la Garza Ramos et al, 2022, United States11 | Spinal Neurosurgery | ML (ANN) | Prediction of Perioperative Blood Transfusion: Accuracy: 77% Sensitivity: 80% |
| Azimi et al, 2014, Iran12 | Spinal Neurosurgery | ML (ANN) | Surgical satisfaction Accuracy: 96.9% |
| Elahian et al, 2017, United States13 | Epilepsy and Functional Neurosurgery | ML (Logistic regression) | Abnormal SOZ identification Accuracy: 83% |
| Roy et al, 2020, United Kingdom14 | Epilepsy and Functional Neurosurgery | ML (k-NN, SGD, XGBoost, and CNN) | Seizure-wise cross validation Accuracy: 90.1% Patient-wise cross validation Accuracy: 56.1% |
| Saputro et al, 2019, Indonesia15 | Epilepsy and Functional Neurosurgery | ML (SVM) | Classification of Seizure Type: Accuracy: 91.4% Sensitivity: 90.25% Specificity: 97.83% |
| Ahmedt et al, 2018, Australia16 | Epilepsy and Functional Neurosurgery | DL (CNN, Long short-term memory) | Multi-fold cross-validation Accuracy: 92.10% Leave-one-subject-out cross-validation Accuracy: 58.49% |
| Varatharajah et al, 2022, United States17 | Epilepsy and Functional Neurosurgery | ML (Naïve Bayes classifier) | Prediction of Seizure Occurrence 1 year Post-op: Dataset 1 Accuracy: 78% Dataset 2 Accuracy: 76% |
| Kassahun et al, 2014, Germany18 | Epilepsy and Functional Neurosurgery | ML (Genetic -based data mining, ontology-based classification) | Epilepsy Classification Accuracy: 60% |
| Shi Z et al, 2020, China19 | Endovascular and Cerebrovascular Neurosurgery | ML (CNN) | Aneurysm detection/Lesion level: Accuracy: 88.6% Sensitivity: 94.4% Specificity: 83.9% |
| Faron et al, 2020, Germany20 | Endovascular and Cerebrovascular Neurosurgery | ML (CNN) | 1st diagnosis sensitivity: 95% 2nd diagnosis sensitivity: 94% |
| Park et al, 2019, United States21 | Endovascular and Cerebrovascular Neurosurgery | DL (DNN) | Threshold of Aneurysm Size for Intraprocedural Rupture Accuracy: 68.7% Sensitivity:60% Specificity: 79.1% |
| Nishi et al, 2021, Japan22 | Endovascular and Cerebrovascular Neurosurgery | DL (CNN) | Subarachnoid Hemorrhage Detection: Patient based analysis: sensitivity: 99% Specificity: 92% Slice based analysis: Sensitivity: 89% Specificity: 98% |
| Cepeda et al, 2021, Spain23 | Neurosurgical Oncology | DL (Inception V3, Cox regression) | B-mode Accuracy: 72–89% Elastography Accuracy: 79–95% |
| Tandel et al, 2020, India24 | Neurosurgical Oncology | DL (CNN) and ML (CNN) | Classification between normal and abnormal (tumorous) Accuracy: DL: 94.7% ML: 73.1% |
| Patil et al, 2023, India25 | Neurosurgical Oncology | DL (Ensemble deep-CNN) | Classification of early stage brain tumor Accuracy: 97.77% |
| Alnowami et al, 2022, Saudi Arabia26 | Neurosurgical Oncology | DL (DenseNet) | Ten-fold cross-validation; Accuracy: 96.52% Sensitivity: 98.5% Specificity: 82.1% |
| Khan et al, 2021, United Kingdom27 | Neurosurgical Oncology | ML (DNN) | Surgical phase accuracy: 91% Surgical Steps accuracy: 76% |
| Park, Y.W. et al, 2021, South Korea28 | Neurosurgical Oncology | ML (Radiomics) | Differentiating GBM recurrence from Radiation Necrosis RN post-concurrent chemoradiotherapy: Accuracy: 78% Sensitivity: 66.7% Specificity: 87% |
AI, artificial intelligence; DL, deep learning; ML, machine learning; CNN, convolutional neural network; XG, extreme gradient; DNN, deep neural network; ANN, artificial neural network; SGD, stochastic gradient descent; SVM, support vector machines; DenseNet, densely connected convolutional network; SSI, surgical site infection; SOZ, seizure onset zone; GBM, glioblastoma multiformes.