Applications of open-source software ROAST in clinical studies: A review.

BACKGROUND: Transcranial electrical stimulation (TES) is broadly investigated as a therapeutic technique for a wide range of neurological disorders. The electric fields induced by TES in the brain can be estimated by computational models. A realistic and volumetric approach to simulate TES (ROAST) has been recently released as an open-source software package and has been widely used in TES research and its clinical applications. Rigor and reproducibility of TES studies have recently become a concern, especially in the context of computational modeling.
METHODS: Here we reviewed 94 clinical TES studies that leveraged ROAST for computational modeling. When reviewing each study, we pay attention to details related to the rigor and reproducibility as defined by the locations of stimulation electrodes and the dose of stimulating current. Specifically, we compared across studies the electrode montages, stimulated brain areas, achieved electric field strength, and the relations between modeled electric field and clinical outcomes.
RESULTS: We found that over 1800 individual heads have been modeled by ROAST for more than 30 different clinical applications. Similar electric field intensities were found to be reproducible by ROAST across different studies at the same brain area under same or similar stimulation montages.
CONCLUSION: This article reviews the use cases of ROAST and provides an overview of how ROAST has been leveraged to enhance the rigor and reproducibility of TES research and its applications.

Introduction

Transcranial electrical stimulation (TES) has been broadly investigated as a therapeutic technique for a wide range of neurological disorders such as major depression [1], epilepsy [2-5], Parkinson’s disease [6], chronic pain [7,8], and stroke [9]. For more systematic reviews, see Refs. [10,11]. The location of stimulation electrodes on the scalp and the exact dose of stimulating current contribute to the rigor and reproducibility of TES studies, as these factors directly determine the stimulation intensity and focality at the desired targets in the brain [12]. Computational models have been heavily used for estimating electric field distribution in each individual head [13-15]. However, these models are not readily accessible to medical doctors. Since the introduction of MRI-derived (i.e., individualized) models [13] and model validation [16], the use of current-flow models has greatly expanded to increase the study rigor (Fig. 1). However, proprietary engineering modeling tools (e.g., COMSOL, Abaqus) are technically sophisticated and difficult to implement for most medical doctors [13-15,17]. Open-source software usually have a steep learning curve for researchers without a solid background in computer science (e.g., SciRun, [18]. We recently released a realistic and volumetric approach to simulate TES (ROAST) which succeeds in terms of automation, ease-of-use, speed, and experimental validation [19]. Compared to the other major open-source software in the field, SimNIBS [15,20], ROAST advocates volumetric and realistic modeling of the anatomy in the head tissues and performed on par with SimNIBS when tested out-of-box on validation data [19,21].

Fig. 1.

Number of publications in PubMed returned by searching “computational models transcranial electrical stimulation”. Major open-source software for TES modeling are noted at their time of release. Note the release time of the software may be earlier than the time of their corresponding publication.

As a new software in the field of TES research, ROAST has gained hundreds of users in a short period of time (Fig. 2). It has been used to model over 1800 individual heads spanning across 12 applications (Table 1). By ensuring the accuracy and replicability throughout the entire modeling process including head segmentation, electrode location and placement, and dose of the stimulation, ROAST helped enhance the rigor and reproducibility of TES studies. Various montages were modeled and electric field magnitudes at the same brain areas under similar montages were reproducible across different studies (Table 2). This paper reviews the adoptions of this software and the use cases in detail, in the hope that future TES research and applications can have a reference on how to leverage readily available computational models to enhance rigor and reproducibility.

Fig. 2.

Traffic data from Google Analytics for the hosting website of ROAST. (A) Daily downloads since the first release (V1.0). Time points of major version upgrades are noted by vertical gray lines. Note that traffic data are not available immediately after V1.0 as we did not set up traffic tracking until February 2018. (B) Geographical distributions of visitors.

Table 1

Clinical studies that used ROAST to model individual heads under different research contexts.

Applications	Number of Subjects Modeled (References)	Use Purposes
Aging effects	N = 587 [22]	(I), (III)
	N = 130 [23]	(I), (II), (V)
	N = 54 [24]	(I), (II), (III)
Alzheimer/Dementia	N = 2 [25]	(II), (III)
	N = 60 [26]	(II), (III), (VI)
Brain tumor/lesion	N = 2 [27]	(I), (II)
	N = 2 [28]	(II), (VI)
	N = 8 [29]	(I), (II), (VI)
Cerebellar stimulation	N = 4 [30]	(I), (II), (VI)
	N = 12 [31]	(I), (II), (VI)
	N = 18 [32]	(I), (III), (IV)
	N = 10 [33]	(I), (III), (IV)
	N = 12 [34]	(I), (III), (IV)
	N = 25 [35]	(I), (III), (IV)
Cognition	N = 16 [36]	(I), (II), (VI)
Depression	N = 151 [37]	(I)
Epilepsy	N = 2 [38]	(I)
	N = 12 [39]	(I), (II), (VI)
Functional connectivity	N = 10 [40]	(I), (II)
Inter-individual variability	N = 57 [41]	(I), (II), (IV), (V) (VI)
	N = 50 [42]	(I), (II), (V), (VI)
	N = 14 [43]	(I), (II), (VI)
	N = 2 [44]	(I), (IV)
	N = 32 [45]	(II)
	N = 47 [46]	(I)
	N = 60 [47]	(I)
	N = 240 [48]	(I), (II), (III), (VI)
	N = 29 [49]	(I), (V)
	N = 47 [50]	(I), (V)
	N = 15 [51]	(II), (VI)
	N = 90 [52]	(II), (V), (VI)
Schizophrenia	N = 21 [53]	(I), (II), (VI)
	N = 17 [54]	(I)
Substance use disorder	N = 5 [55,56]	(II), (IV), (V), (VI)
Working memory and attention	N = 15 [57]	(I), (II)
Total	N = 1858

Use purposes include: (I) ROI analysis of E-field against clinical outcomes; (II) Visualization of the E-field at ROI; (III) Voxel-based morphometry; (IV) Optimization of the stimulation; (V) Dose control; (VI) Visualization of electrode placement.

Table 2

Details in the studies reported in Table 1. Electrode names follow international 10/05 convention unless otherwise specified.

Number of Subjects Modeled (References)	Electrode montage (high-definition (H) or conventional(C))	Which brain area is specifically studied?	E-field or current density output by ROAST at studied brain area (normalized to 1 mA stimulation)	E-field correlates with the clinical outcome?	Patients or healthy subjects?
N = 587 [22]	F3-F4 & C3-Fp2 (C)	Entire brain	Average median were 0.007 A/m2 and 0.009 A/m2 for F3-F4, and 0.011 A/m2 and 0.012 A/m2 for C3-Fp2 montage in the older and young adult cohort, respectively.	E-field inversely correlated with brain atrophy	Healthy old and young adults
N = 130 [23]	F3-F4 (C)	White matter hyperintensities (WMH)	WMH regions had a maximum of 1.77 V/m.	Changes in E-field positively correlated with the total lesion volume.	Healthy old adults
N = 54 [24]	F3 (C)	Left M1 and DLPFC	N/A	E-field decreased with scalp-to-cortex distance in mild cognitive impairment converters.	Normal aging and mild cognitive impairment converters
N = 2 [25]	F3-F4 (C)	Frontal cortex	Peak E-field of 0.3 V/m.	N/A	Patients with early stage Alzheimer’s disease
N = 60 [26]	FT7-AF8 (C)	Left anterior/middle temporal lobe	N/A	N/A	Patients with dementia
N = 2 [27]	Anterior-posterior and left-right array (H)	Brain tumor	Average E-field at tumor is 0.17 V/m.	Presence of peritumoral edema resulted in decreased E-field magnitude within the tumor.	Patients with brain tumor
N = 2 [28]	F3-F4, P3-P4 (C&H)	Cortical surface	Peak E-field of 0.16 V/m.	N/A	Healthy and patient with multiple sclerosis
N = 8 [29]	C3-FP1 (C)	Left M1	Average E-field is 0.12 ± 0.03 V/m (range 0.08–0.17 V/m)	E-field magnitude applied to the left M1 correlated with changes in global connectivity of the right M1.	Patients with left-sided glioma
N = 4 [30]	E133-E18 in EGI HCGSN-256 system (C); anode Iz - cathodes Oz, O2, P8, PO8 (H)	Cerebellum	0.2 V/m - 0.25 V/m under montage E133-E18; Average 0.1 V/m under montage anode Iz - cathodes Oz, O2, P8, PO8	Amplitude and orientation of E-field is related to bursting and complex spiking in Purkinje cells in the cerebellum.	Healthy subjects
N = 12 [31]	PO9h - PO10h Exx7 - Exx8 (H)	Cerebellum	Peak E-field of 0.15 V/m.	Mean E-field strength was a good predictor of the latent variables of oxy-hemoglobin (O2Hb) concentrations and log10-transformed EEG bandpower.	Patients with hemiparetic chronic stroke
N = 18 [32]	Celnik montage (C)	Cerebellum	Peak E-field of 0.15 V/m.	E-Field increased significantly at the targeted cerebellar hemisphere at an old age.	Healthy subjects
N = 10 [33]	PO9h-PO10h Exx7-Exx8 (H)	Cerebellum	Average ~0.04 V/m.	A linear relationship between successful functional reach in post-stroke balance rehabilitation and E-field strength was found.	Patients with chronic stroke
N = 12 [34]	PO9h-PO10h Exx7-Exx8 (H)	Cerebellum	Average ~0.05 V/m.	The changes in the quantitative gait parameters were found to be correlated to the mean E-field strength in the cerebellar lobules.	Patients with chronic stroke
N = 25 [35]	I1-Exx25 (C)	Cerebellum	N/A	tDCS-related metabolite changes may be related to the strength of the E-field induced at the region of interest.	Healthy subjects
N = 16 [36]	CP5-CZ TP7-TP8 (C)	Lexical (ventral) and sublexical (dorsal) pathways for language	Average ~0.04 A/m2.	Sub-lexical proficiency is associated with greater effects of tDCS stimulation.	Healthy subjects
N = 151 [37]	C2-FT8 (H)	Left amygdala and left hippocampus	Average ~0.11 V/m.	High electrical fields are strongly associated with robust volume changes in a dose-dependent fashion.	Patients with depression
N = 2 [38]	Left and right earlobes and infra-auricular (H)	Deep brain sampled by sEEG electrodes	Maximum of 0.4 V/m.	E-fields measured in vivo are highly correlated with the predicted ones.	Patients with epilepsy
N = 12 [39]	Various montages such as T8, Oz - T7 (H)	Deep brain sampled by sEEG electrodes	Maximum of 0.5 V/m.	E-fields measured in vivo are highly correlated with the predicted ones.	Patients with epilepsy
N = 10 [40]	PO7, PO3 - Cz (H)	Motion area	Average E-field magnitude on the left motion area is 0.16 V/m, and on the right motion area 0.09 V/m.	Functional connectivity (between motion area and any other region of interest) increases in proportion to the E-field strength in the region of interest.	Healthy subjects
N = 57 [41]	Cz-Oz (C)	Entire brain	Average E-field is 0.13 ± 0.05 V/m (min = 0.08 V/m, max = 0.36 V/m).	Variability of power increase in alpha-oscillations was significantly predicted by E-field from individual modeling.	Healthy subjects
N = 50 [42]	Directional montage: CP5-FC1 (H); Conventional montage: C3-FP2 (H)	Left M1	Directional montage: 0.19 ± 0.04 V/m; Conventional montage: 0.18 ± 0.04 V/m.	Fixed-dose tDCS yields substantially variable E-field intensities in left M1 due to inter-individual variability.	Healthy subjects
N = 14 [43]	F3-F4 (C)	Entire brain	N/A	Median E-field in brain regions near the electrodes were positively related to tDCS intervention responses.	Healthy older adults
N = 2 [44]	Fp2-CCP3 (H) Exx20-FFT7h or F7h (H)	M1 Broca’s area (BA44)	Fp2-CCP3: 0.16 V/m.	Lesions that were larger, closer to the ROI, and had a higher conductance tended to have the greatest impact on E-field magnitude.	Healthy subjects, with lesions added in the model
N = 32 [45]	AF3-CP5 (C)	Entire brain	N/A	N/A	Healthy subjects
N = 47 [46]	F3-F4 (C)	Inferior frontal gyrus	Median of 0.047 V/m.	Including E-field in the regressions did not change the effect of tDCS.	Healthy subjects
N = 60 [47]	F3-F4 (H)	Frontal cortex	0.06–0.10 V/m.	E-field accounted for 54%−65% of the variance in tACS-related performance improvements.	Healthy old adults
N = 240 [48]	CP5-CZ (C)	Inferior parietal lobule (IPL)	Average ~0.2 mA/m2.	Across all age groups, CSF and gray matter volumes significantly predicted the E-field at the target sites.	Healthy subjects
	F3-FP2 (C)	Middle frontal gyrus (MFG)
N = 29 [49]	Left motor hotspot and left neck (C)	Motor cortex	Average 0.17 V/m.	Transcranial electrical stimulation motor threshold significantly correlated with the ROI-based reverse-calculated tDCS dose determined by E-field modeling.	Healthy subjects
N = 47 [50]	1 cm posterior to F3-F4 (C)	Left prefrontal cortex	N/A	Cortical thickness in left prefrontal cortex correlates with anodal tDCS efficacy.	Healthy subjects
N = 15 [51]	CP5-Cz (C)	Entire brain	Average ~0.14 mA/m2	N/A	Healthy subjects
N = 90 [52]	F3 and the right supraorbital (C)	Left middle frontal gyrus	Average ~0.12 mA/m2	N/A	Healthy subjects
N = 21 [53]	Anode: left DLPFC (between F3 & FP1 Cathode: left TPOJ (between T3 & P3) (C)	TPOJ and auditory association regions	Average ~0.25 V/m.	E-field strength at anterior regions correlated significantly with less robust clinical response.	Patients with schizophrenia
N = 17 [54]	Anode: left DLPFC (between F3 & FP1 Cathode: left TPOJ (between T3 & P3) (C)	Left transverse temporal gyrus	N/A	tDCS responders displayed higher E-field strength in the left transverse temporal gyrus at baseline compared to nonresponders.	Patients with schizophrenia
N = 5 [55,56]	OI2-E145 in EGI HCGSN-256 system (H)	Cerebellum	Average ~0.12 V/m.	N/A	Patients with stroke
N = 15 [57]	F3-F4 (C)	Left DLPFC and left VLPFC	Average median at left DLPFC was 0.0407 A/m2, and at left VLPFC was 0.0265 A/m2.	E-field in the left DLPFC under active stimulation positively correlated with the beta values as measured functional connectivity metrics.	Healthy old adults
N = 1858

N/A: data not reported in the paper. EEG: electroencephalography; CSF: cerebrospinal fluid; tDCS/tACS: transcranial direct/alternating current stimulation; ROI: region of interest; DLPFC/VLPFC: dorso/ventral lateral prefrontal cortex; M1: primary motor cortex; TPOJ: temporo-parietal-occipital junction.

Methods

Literature search

To find out the trend in the literature that utilized modeling for TES research, keywords “computational models transcranial electrical stimulation” were used to search the literature on PubMed. Number of publications by year was returned and plotted.

Adoptions of ROAST

Shortly after the release of ROAST, we have been tracking user downloads on the website that hosts ROAST (https://www.parralab.org/roast/) by Google Analytics. Daily downloads and geographic locations were stored and plotted.

Citation report

All the papers found on Google Scholar that cited ROAST publications [19,58] were reviewed in April 2022. For each paper, we looked up the number of subjects that were modeled by ROAST, the clinical applications of the subjects that were studied, and the purpose of computational modeling in that study.

Rigor and reproducibility

A workshop organized by the National Institute of Mental Health in 2016 discussed major factors contributing to the rigor and reproducibility of TES research [12]. The factors that relate to computational modeling include locations of the placed electrodes on the scalp and the dosing of the stimulation. To show how ROAST helps to enhance rigor and reproducibility in those clinical studies found on Google Scholar that used ROAST for modeling more than one individual head, we extracted the following information and compared them across studies: electrode montage and electrode type (conventional (C) vs. high-definition (HD)), stimulated brain areas, achieved intensity of electric field at the stimulated areas (normalized to 1 mA dose), correlation between modeled electric field and clinical outcomes, and subject characteristics (patients vs. healthy).

Results

Computational models of TES tend to be widely adopted

It is obvious that more and more TES studies start to use computational models (Fig. 1), especially since the introduction of individualized modeling from MRIs [13]. SimNIBS, SciRun, and ROAST all helped push the adoption of current-flow models in the literature. Specifically, ROAST has been downloaded 1598 times (1414 unique downloads; see Fig. 2) by April 2022.

ROAST has been heavily used for individualized TES modeling

According to Google Scholar, the papers in which ROAST was published [19,58] had been cited 225 times by April 2022. Among these, 15 are dissertations and 24 are reviews and book chapters. We reviewed the remaining 186 papers, and found 94 clinical TES studies that used ROAST for computational modeling. Table 1 summarizes all the results for each specific clinical application. As a reference, note that SimNIBS [15,20] has been cited over 800 times, and SciRun for TES simulation [18] has been cited 57 times. One of the studies in Table 1 also used SimNIBS to model the 32 heads but did not find any significant difference in predicted electric field compared to ROAST [45].

It is clear from Table 1 that ROAST has been applied in clinical studies spanning across 12 applications and modeled 1858 individual heads, thanks to its scripting feature that allows easy batch processing. Most of these studies used ROAST to visualize the stimulation electrodes and the electric field distribution at the region of interests (ROI), and to correlate the simulated electric field intensities at the ROIs with clinical outcomes. Some of these studies used ROAST to calculate the dosing of stimulation, optimize the stimulation montage, or perform voxel-based morphometry using the generated tissue segmentation. The study that modeled the most subjects was [22]; where N = 587 healthy older adults under TES were modeled. The results showed that the amount of stimulation current that reaches the brain decreases with increasing atrophy, suggesting that adjusting current dose in older adults based on degree of atrophy may be necessary to achieve desired stimulation benefits. It was not possible to perform TES modeling studies with rigor and reproducibility for over 500 subjects before ROAST was created, as one had to run head segmentation, electrode placement, and electric field computation by hand in various software [13,17,59], where uncertainties may be introduced by manual operations of these software in the modeling process. Other representative studies include: Ref. [26] simulated N = 60 dementia patients to correlate the model-predicted electric field at ROIs with clinical data to evaluate the therapeutic efficacy of a multi-day TES regime on language impairment in patients with semantic dementia. Ref. [29] used ROAST to model N = 8 glioma patients in their study of TES feasibility on these patients. They showed that patient-specific modeling of electric field in the presence of tumor may contribute to understanding the dose-response relationship of this intervention. Ref. [32] modeled N = 18 subjects at different ages for cerebellar transcranial direct current stimulation and found that cerebellar shrinkage and increasing thickness of the highly conductive CSF during healthy aging can lead to the dispersion of the current away from the lobules underlying the active electrode. Ref. [36] built individualized models for N = 16 subjects to help determine the best montage for selective modulation of dorsal and ventral pathways of reading in bilinguals. Ref. [37] used ROAST to calculate the electric field intensities in N = 151 patients with severe depression undergoing electroconvulsive therapy (ECT) and found that the electric fields predicted by ROAST positively correlate with the volumetric changes of the brain due to ECT. Ref. [39] compared in vivo measured electric fields during TES on N = 12 epilepsy patients with their individual models generated by ROAST to validate the models. Ref. [40] built N = 10 individualized models using ROAST to study if electric field intensities at the ROIs positively correlate with functional connectivity. Another relatively large study [48] leveraged ROAST to model N = 240 individuals to study the effects of cortical anatomical parameters such as volumes, dimension, and torque on simulated TES current density in healthy young, middle-aged, and older males and females. Ref. [53] modeled N = 21 individual heads to assess the target engagement in their study of TES on antipsychotic-resistant auditory verbal hallucinations in schizophrenia. Refs. [55,56] built individualized head models for N = 5 subjects to compute the optimal electrode montage to target the cortico-cerebello-thalamo-cortical loop for improving substance use disorder. Ref. [57] modeled N = 15 subjects to predict significant changes of functional connectivity observed in the working memory network from an acute TES application.

In addition, many studies run the models on the example head included with ROAST or an individual sample from the investigators. These work cover various clinical applications including: attention-deficit hyperactivity disorder [61,62], aging [63], associative memory [64,65], attention [66-68], body awareness [69], cognitive control and function [70]; Fusco et al. [125]; [71,72], connectivity [73], decision making [74-77]; Schulreich and Schwabe [126], declarative learning [78], depressive disorder [79], electroencephalography (EEG) research [80-83], imitation [84], memory retrieval [85-87], mind wandering [88,89], motor learning [90-95], motor skills [96-98], neurorehabilitation [60], neurovascular coupling [99], obsessive-compulsive disorder [100], phantom limb pain [101], post-anoxic leukoencephalopathy [102], reading speed [103], schizophrenia [104], social anxiety disorder [105], stroke [106], visual perception [107,108], and working memory [109-115].

Note that for those studies that involved subjects with pathological head anatomies (e.g., tumor or lesion), customized segmentation was performed and integrated into the ROAST pipeline to account for these anatomies [25,29]. This is because the segmentation function in ROAST [116] was developed for normal head anatomy only.

ROAST helps to enhance the rigor and reproducibility

From Table 2, we can see that ROAST has been used to model various electrode montages to stimulate different brain areas. 29 out of the 35 studies in Table 2 used bipolar montages, and 21 of these bipolar montages are conventional pad electrodes. Most of the studies in Table 2 were interested in stimulating the primary motor cortex (M1), frontal cortex and cerebellum. For the primary motor cortex, Ref. [29] used bipolar montage C3-FP1 with conventional electrodes and achieved an average electric field of 0.12 V/m at the left M1 with 1 mA stimulating current. Ref. [42] obtained an average of 0.19 V/m under montage CP5-FC1 with high-definition electrodes, and 0.18 V/m under montage C3-FP2. Ref. [44] achieved 0.16 V/m averaged electric field with high-definition electrodes Fp2-CCP3. For the frontal cortex, Ref. [25] obtained a peak electric field of 0.3 V/m with montage F3–F4 using conventional electrodes. With the same montage, Ref. [46] achieved a median electric field of 0.047 V/m at inferior frontal gyrus. Also with the same montage but high-definition electrodes, Ref. [47] showed an electric field in the range of 0.06–0.10 V/m in the frontal cortex. With montage F3 and the right supraorbital, Ref. [52] outputs an average current density of 0.12 mA/m2 at the left middle frontal gyrus. For the cerebellum, both [33,34] report an average of about 0.05 V/m under the same montage of PO9h–PO10h using high-definition electrodes. These results suggest that ROAST may help to enhance the rigor of TES models as similar electric field intensities were reproducible across different studies at the same brain area under same or similar stimulation montages.

In Table 2, 21 out of the 35 studies focus on healthy subjects including old and young adults. The other 14 studies in Table 2 build models for patients with the corresponding clinical applications in Table 1. For all the studies in Table 1 with Use Purpose (I), i.e., ROI analysis of E-field against clinical outcomes, we noted in Table 2 the detailed correlation between the predicted electric field and the studied clinical outcome/metric. Except one study [46], all the other studies in Table 2 report significant correlations between the electric field intensity and the outcome of stimulation or the inter-individual variability.

Discussions and conclusions

It is clear that computational models are becoming more and more intensively used in the research and clinical applications of TES to enhance rigor and reproducibility. As a new modeling tool in the TES community, ROAST can be improved in several ways to further strengthen study rigor and reproducibility: (1) ROI analysis: a function that allows users to automatically read out electric fields at the ROIs either in the individual head or the standard head space [117]. (2) Interface with other open-source software. For example, researchers in source imaging using electroencephalography/magnetoencephalography (EEG/MEG) rely on the same forward models that ROAST generates [118]. We have developed an interface [119] that allows users to import the models of a standard head from ROAST into Brainstorm, a popular software for EEG/MEG source localization [120]. (3) Interface of customized segmentation. This will allow users to add additional, customized geometry in the model. (4) Integration of modern deep-learning engine for segmentation of pathological head anatomies mostly presented in clinical populations [121]. This will significantly expand the clinical adoptions of this software, as the conventional segmentation algorithm used by ROAST [116] is not capable of handling pathological heads. (5) Development of a platform that allows calibration of tissue conductivities for more accurate and personalized modeling. TES models overestimate the electric field compared to intracranial electrical recordings [16], but underestimate the magnetic field induced by the stimulation current compared to actual measurements [122]. Future work will leverage state-of-the-art recording techniques such as in-vivo stereotactic EEG electrodes inserted into the deep brain [123], or in-vivo imaging of magnetic fields in the head induced by the stimulation current [124] to calibrate the models and derive individualized tissue conductivities. This will facilitate more precise dosing and spatial targeting for the stimulation.

In conclusion, the era of precise medicine has come including clinical applications of TES where highly individualized and accurate computational models are becoming more readily accessible with constantly improved software and computational power.