Table 2.

Summary of different approaches for ICP estimation and MLS quantification.

Ref.	ML/DL	Algorithm/Method	Dataset Size	TBI-related Clinical Assessment	Performance	Main Contribution
[77]	ML	ReliefF feature selection, SVM	17	ICP > 15 vs. ICP ≤ 15	Acc: 81.79 ± 2.3 Sen: 82.25 ± 1.7 Spec: 81.20 ± 0.04	Textural-based ICP estimation using Histogram analysis, GLRLM, DWPT, Fourier Transform, and DT-CWT
[80]	ML	SVM	17I	ICP > 12 vs. ICP ≤ 12	Acc: 70.2 ± 4.5 Sen: 65.2 ± 8.6 Spec: 73.7 ± 4.6	Textural-based ICP estimation using Histogram analysis, GLRLM, DWPT, Fourier Transform, DT-CWT as well as hematoma volume, manual MLS measurement, age, and ISS
[81]	ML`	SVM	17	ICP > 12 vs. ICP ≤ 12	Acc: 73.7 Sen: 68.6 Spec: 76.6	Textural-based; improving previous model [80] by adding intracranial air cavities, ventricle size related feature
[83]	ML	GA-SVM and GA-KNN based feature selection, SVM classification method	59	ICP > 15 vs. ICP < 15	Acc: 86.5	Textural-based; using anisotropic complex wavelet as textural feature for ICP classification
[82]	ML	GA-SVM based feature selection, SVR classification method	59	ICP > 15 vs. ICP < 15	Acc: 0.94 MAE: 4.25 mmHg	Textural-based; comparing anisotropic complex wavelet transform extracted features vs. DT-CWT extracted features in ICP classification.
[76]		Hounsfield units thresholding	20	ICP > 20 vs. ICP < 20	Acc: 0.67	Morphological-based; brain parenchyma segmentation, ICP estimation based on csfv/icvv ratio
[69]	ML	Black box model	11	ICP	MAE: 4.0 ± 1.8 mmHg	Physiological-based; non-invasive prediction of ICP using ABP and FV of major cerebral arteries.
[70]	ML	SVM	446	ICP	Mean ICP error: 6.7 mmHg	Physiological-based; noninvasive measurement of ICP using ABP and FV of major cerebral arteries
[71]	ML	Linear regression	74	ICP > 20 vs. ICP ≤ 20	AUC: 0.94	Morphological-based; estimating the probability of increase ICP by measuring MRI-based ONSD
[80]	ML	Gaussian mixture model	57	ICP > 12 vs. ICP ≤ 12	Acc: 0.70	Textural-based ICP estimation; using tissue textural features, manual MLS quantification, ISS, and age
[79]	ML	Information gain ratio, GA feature selection, SVM	57	ICP > 12 vs. ICP ≤ 12	Acc: 0.70 Sen: 0.65 Spec: 0.73	Textural-based ICP estimation; automated iML detection; using textural features, MLS, and blood amount to estimate ICP
[87]		Bezier curve, GA	81	MLS	Acc: 0.95	Symmetry-based; detecting dML at the foramen of Monro level; Low performance in case of severe TBI
[88]	ML	H-MLS (Linear Regression based), visual symmetry information	11	MLS	-	Symmetry-based; tracing dML based on the brain hemorrhage detection
[89]		Weighted midline, maximum distance	41	MLS	Acc: 0.92	Symmetry-based; estimating MLS based on the maximum distance between WML and iML close to the foramen of Monro
[94]	ML	CT density, spatial filtering, cluster analysis	273	MLS	Sen:1 Spec: 0.98	Landmark-based; using spatial filtering, CT density thresholds, and cluster analysis to segment blood and CSF. The symmetry of CSF pixels in the lateral ventricles is used to assess MLS
[90]	ML	Level-set	170	MLS	Acc: 0.68	Landmark-based; automating CT slice selection, rotation, and segmentation
[91]	ML	Active contour, Logistic regression	48	MSS vs. MLS	AUC: 0.71 Acc: 0.79	Landmark-based; volumetric measurement of displaced brain mass was significantly correlated with GOS on discharge
[44]	DL	NPL	313, 318 (Qure25k dataset); 491 (CQ500 dataset) as validation set	MLS	AUC: 0·969 Sen: 0.938 Spec: 0.907	Detecting mass effect which consists of MLS, ventricular effacement, herniation, or local mass effect
[92]	DL	RLDN	189 (CQ500 dataset [44] and local resources)	MLS	F1-score: 0.78	Developing a multi-scale bidirectional FCN based method [95] for midline delineation in severe brain deformation
[93]	DL	U-Net	45	MLS	Acc: 0.94	Deformed right and left hemispheres were automatically segmented. The junction of these two segments was then traced to forms dML.
[43]	DL	U-Net based FCN	38	MLS < 5 mm vs. MLS > 5 mm	Acc: 0.89	MLS estimation at all levels between the lateral ventricles roof and foramen of Monro was performed. The greatest MLS score was considered the final MLS.