Multiple sclerosis (MS) is a chronic, white and gray matter disease of the central nervous system. Gray matter disease in MS is poorly visualized in conventional MRI but has been increasingly studied in recent years using high strength magnets ( de Graaf et al., 2013). The use of MRI in tracking disease of the human brain and spinal cord in patients with MS is central to the diagnosis and treatment of the disease.

Development of computational models for patient-specific requirements based on human pathophysiology individualized to patient-specific data is needed as we move forward with advanced techniques such as 3D printing in medicine. For starters, the potential to improve diagnosis and optimize clinical treatment by predicting outcomes of therapies is attainable. For instance, the accurate prediction of rupture of abdominal aortic aneurysm is possible through patient-based diagnostic tools coupled to medical imaging ( Ricotta et al., 2008). However, most results might not apply directly to individual patients yet because they are based on averages ( Kent & Hayward, 2007). As an alternative, patient-specific modeling (PSM) can be used as an analytical tool to optimize an individual's therapy. Our study could potentially be useful in building a platform for patient-specific treatment options based on 3D analysis of brain disease, particularly in acute settings such as stroke, mass effect of tumors, midline shift in patients with acute intracerebral hemorrhage, among others.

With rapid strides made in computer-based technologies, brain atlases are ‘constructed’ by computers. This enables such atlases to become plastic or deformable to fit the size/shape of individual brains. To construct brain atlases, collections of micrographs or schematic drawings of brain sections from one or a few brains are used in which anatomical structures such as nuclei, cortical ribbon or tracts, are identified ( Roland & Zilles, 1994). To make assumptions about localization of function and structure at both the macroscopic and microscopic levels, computerized brain atlases are needed. Computerized brain atlases are also used for topographically defined data from the literature ( Roland & Zilles, 1994). The spatial resolution is about 1 mm for structural imaging and is below the cellular scale ( Roland & Zilles, 1994). For understanding the interaction between brain areas and regions, subcortical nuclei, gyri and sulci, the resolution appears to be sufficient ( Toga et al., 2006).

Image segmentation is crucial in medical image analysis and is perhaps the most critical step in many clinical applications ( Despotović et al., 2015). In brain MRI analysis, image segmentation is used for measuring and visualizing the brain’s anatomical structures, analyzing changes and identification of pathological regions, as well as for surgical planning and image-guided interventions. Recent advances in brain MRI have provided large amount of data with an increasingly high level of quality but analysis of large and complex MRI datasets is onerous for clinicians, who still extract information manually. Since errors due to inter- or intra-operator variability studies rack up when manual analyses are done, brain MRI data analysis requires inventions in computerized methods to improve disease diagnosis. Increasingly, computerized methods for MR image segmentation, registration, and visualization have been extensively used to assist doctors in qualitative diagnosis ( Despotović et al., 2015). It is to be noted specifically that segmentation in 2D is a dated technique; no reconstruction of ‘burden of disease’ in MS as depicted in a 3D printed brain format has been attempted. Indeed, there are no publications that describe the presentation of a one’s own brain in a 3D model, depicting the extent of that individual’s disease. While segmentation of MRI lesions of brain is not new, collation of 2D data into a printable 3D model, is novel.

To help the patient understand the extent of the disease is probably cathartic and revealing although each individual patient may react differently. The primary goal of our endeavor is to educate patient(s) and physician (s) alike regarding the magnitude of a medical disease and the immediacy of treating such a ravaged brain. Our concept borrows from the design and modeling of normal, anatomically-detailed, 3D representations of the normal male and female human bodies and acquisition of transverse CT, MR and cryosection images of representative male and female cadavers in the Visible Human Project.

From a patient’s perspective, holding one’s own brain in the palm of a hand delves into a hitherto unknown and previously unexplored dimensions. Looking en face at the disease, particularly for a condition that has minimal or no surgical options probably gives patients a better perspective about their disease, but could also evoke fear. With 3D modeling, we enter a novel but untouched world in disease presentation to patients. Only time can tell if more patients will embrace such an idea and wish to explore the unknown.

We obtained routine MRI images of the brain from a young Caucasian woman in her early 20s who came to our neurology clinic for the first time following a hospital visit for headache, mild gait ataxia and visual impairment in her right eye that she had developed over the two days prior to presentation. Her diagnosis of MS was established based on the following MRI of brain findings: T1-gadolinium enhancing lesions and T2 lesions present at the time of diagnosis, in multiple areas of the brain (periventricular, juxtacortical, infratentorial areas), supporting McDonald’s criteria (2010), and her history/clinical examination findings. As well, mimics were excluded and no requirement for lumbar puncture or additional testing were needed. Her MRI images (Phillips 3T TX, software 3.2 version) had the following parameters: Sag T1 SE 5 Thick x 1 gap DWI 5 Thick x 1 gap, Axial FLAIR 5 Thick x 1 gap, Axial T1 SE 5 Thick x 1 gap, Axial PD 5 Thick x 1 gap, Sag FLAIR (reconstructed to Sagittal, Coronal, and Axial 1.0 mm thick x 0 gap, and Sagittal 3D T1 FFE (reconstructed to Sagittal, Coronal, Axial 1.0 mm thick x 0 gap), respectively. The MRI images showed typical white matter lesions that raised concern for MS; her diagnosis was established after ruling out mimics. Since her brain contained an unusually high lesion load, we opted to print a 3D model to fully ascertain the extent of white matter involvement by total lesion volume. We chose T2 FLAIR lesions to compute lesion load and manually identified lesions within each 1 mm slice of the MRI scan in sagittal, coronal and axial planes, respectively. The total combined lesion load was 95,774 mm ³, suggesting axonal transection in this volume of brain tissue. A seminal publication ( Trapp et al., 1998) showed that active MS lesions, defined on a histological basis, had 11,236 transected axons per mm ³ of tissue. This underscores the importance of the burden of disease and the therapeutic challenges that accompany repairing each mm ³ of tissue lost to disease. Our patient had a total white matter lesion load of 95,774 mm ³ corresponding to a loss of 10 ⁹ axons. Since no study had characterized a patient’s total lesion volume loss in 3D in MS, comparison of our results to any published literature is not possible.

The 3D modeling software used in the analysis of our patient is an FDA-cleared proprietary product designed specifically for generating 3D models from MRI data. It includes functionality to overlay the 3D reconstructions onto the cross-sectional MRI slices in order to verify the accuracy of volumes. The software is also used routinely to reconstruct 3D models for pre-surgical planning and design of custom devices for clinical care by the parent company of this software, Materialise NV, Belgium. It is a dedicated medical image processing software designed for image segmentation. This patient presented with de novo MS. The lesions, as described, are T2 lesions and were volumetrically assessed. Prior to the publication, one of us (JA) wrote to Dr Trapp and he agreed that any new T2 lesion or T1 gadolinium enhanced lesion would represent a ‘new lesion’ in a patient who presents for the first time. Acute lesional volumetric analysis ( Trapp et al., 1998) shows that 1 cu mm3 destroys 11 K axons, from axonal transection.

We emphasize that our model ( Figure 3) is primarily educational but can be modified to document the progression or regression of lesions over time. As well, quantification of T1 black hole volume loss, particularly with the development of automated algorithms, is possible ( Datta et al., 2006). Hopefully, our work will trigger research into the study of regional/global atrophy, focal/total cortical thickness assessment and deep gray matter changes in 3D, a field that is increasingly coming to light in conventional studies using Structural Image Evaluation Using Normalization of Atrophy software and statistical parametric mapping analysis ( Pagani et al., 2005). Additionally, a platform to document changes accurately using computer-assisted automated algorithms that are universally accepted and standardized will be developed. This is critical given the recent EPIC study findings that showed a disappointing trend in how disease-modifying drugs fail to arrest or impact disability in MS patients ( Cree et al., 2016) since no drug, if any, affects atrophy measures in a meaningful way. For longitudinal studies, it is crucial that research methods are automated, validated, universally accepted, standardized and based on computer-based image analysis tools that can sift through large data sets. Additional enhancements for our 3D model could include such innovations as Cold Spring Harbor’s G2C interactive normal brain models, funded by the Dana Foundation and Hewlett Foundation, wherein structure/function relationships can be gleaned when a 3D brain with disease is superimposed on an interactive normal 3D brain model giving patients and physicians a new perspective on how different anatomical structures are involved and affected in health and disease. Since no two patients are similar, scan quality can vary but so do their file formats. Yet, if the end goal is improvement in quality patient care, one would want to ensure that the 3D models accurately represent the patient’s anatomy which is what one would expect as 3D technologies continue to evolve.

No standardized approach for segmentation of diseases of the brain, including those that claim automation, is accepted as a gold-standard. Automated segmentation protocols were fundamentally developed to process massive data but to date, no database in the world of MS has defined brain atrophy using such techniques.

Many automated segmentation methods that detect brain lesions have been developed in MS ( Udupa et al., 2001; Wu et al., 2006; Zijdenbos et al., 2002) but no study has been validated for commercial or routine use, nor has the depiction of the impact of lesion load in a 3D printed model been published. If such technology can be developed and transferred to the ICU settings, medical and surgical decisions could perhaps be handled better, particularly in acute neurological disorders that cause rapid clinical changes and worsening mass effect and midline shift following intracerebral bleeding, hydrocephalus or cerebral edema owing to mass effect of tumors. New guidelines could be developed for therapeutic and surgical interventions. Could 3D printing introduce a new angle to how lesion load is defined? Can one visualize 3D printing becoming a teaching, diagnostic and decision-making tool in the ICU setting? We think that to accurately document changes that occur in acute neurological diseases such as hemorrhagic strokes with or without mass effect, or cerebral edema from varied causes, a 3D model would be ideal if not mandatory, particularly if available in real time for decision-making in treatment options and patient education.

Since no radiological markers accurately quantify disability in MS, how does one assess objectively, the effect of disease modifying drugs on MS outcomes research? As technology evolves, a routine CT and MRI scan can probably be converted instantly into a 3D model with the help of automatic segmentation algorithms that could be used to document volumetric changes both global, regional and deep gray matter structures. We hope our study is the first step towards such a goal.

Quantitative analysis of WML in large clinical trials assumes a major role particularly in cerebrovascular disease, diabetes mellitus and Alzheimer’s disease, wherein 30% of patients could have some degree of vascular pathology. In population studies, such as the Cardiovascular Health Study (CHS) or the Rotterdam Scan Study (RSS) WMLs have been shown to be associated with age, clinically silent stroke, higher systolic blood pressure, hypertension, atrial fibrillation, among others ( de Groot et al., 2000; de Groot et al., 2000; Longstreth et al., 1996). An urgent unmet need is the assessment of MRI data of WML load in various disease states that is standardized, automated and followed longitudinally. Hopefully, this study is a first of many such attempts in that evolutionary path moving forward.

Written informed consent was obtained from the patient for the publication of the patient’s details and accompanying images.

The MRI files underlying the 3D model of this patient’s brain have not been included to maintain patient anonymity.

Alternative software packages that are available include Slicer (open source) or Osirix (free demo available) to segment the imaging data, and Meshmixer (open source), a digital CAD software, to prepare the 3D model for printing.