Stephen Sebastyan

5D+ Patient Specific Cardiac Models

A significant part of my research has been on time varying models of the heart. So what do I mean by 5D+?

2D and 3D modelling of cardiac anatomy is quite prevelant. Usually, 3D is created from CT/MR volumetric data or from an electrophysiology catheters. It's also possible with (less resolution) using ultrasound. 2D imaging is usually meant by fluoroscopic radiological visualization

For the most part, these models are static or the movement is treated as an artifact in the imaging modality... And this works fine for anatomy that doesn't move or move much (see my orthopaedic work in the CAMA tab).

The heart is a uniquely complicated organ. It moves, or at least it should be moving (if it's not, find a different doctor). It also has two types of forces driving its movement: respiratory and cardiac (pulse).

To complicate things even more, these movements are at different magnitudes and frequencies (hearts beat usually much faster than your breathing cycle). And usually the people that need heart surgeries don't always have a heart that behave in a consistent and predictable fashion.

Below is an atrial fibrillation patient specific model of a left atrium generated by my computer modelling software that can real-time visualize the cardiac movement based on prior patient data

But if this were not complicated enough, heart surgery has another dimension to the problem. In the above model, we looked at the heart as a mechanical device, but what drives it?

The heart has an electrical component to it. The electrical signals are a critical aspect of the cardiac mechanical behavior. In the model below, I superimposed the voltage activation pattern (ie. electrical activity) onto the dynamic model

This is where the 5D+ comes from. We have 3D position, but also two distinct forces that change over time (making 5D), plus (+) the electrical activity.

One might ask, "Why is this research important?". Cardiac disease is the largest source of death in the world. The initial motivation for this research was to help guide surgical tools to threat cardiac arrhythmias. Arrhythmias are a spectrum of conditions manifesting in the heart beating too rapidly (tachycardia), too slowly (bradycardia), or irregularly.

Arrhythmias can lead to thromboembolic events and congestive heart failure. The consequences can be debilitating or deadly. Current treatments of cardiac arrhythmias range from noninvasive strategies, such as pharmacological therapy (which can strain the kidney and liver), to minimally invasive techniques, such as catheter-based ablation to extensive open-heart surgical techniques

With this in mind, catheter ablation can be a more definitive solution than pharmacological intervention without the significant risks of open surgery. However, the procedure comes with challenges in navigating the heart without direct line of sight. At the moment this is done with a static 3D model, but results of this research show that the approach I have developed is significantly more accurate than the leading manually constructed (and automated) commercial models constructed. It also has significantly less variance. Thus, if a surgeon is trying to ablate and isolate an aberrant signal or excise a tumor, the visualization of my approach will more likely reflect the true anatomy, and in turn, this should yield better surgical outcomes.

Instrument Recovery

part of my PhD research work on developing an approach for catheter position and pose recovery in a distorted field

Motivation & Challenges

EAM Systems can be prone to misalignment (progressive and sponentaneous), which we have colloquially termed "drift" and "shift". In other words, the reference origin can spontaneously or progressive move from it's previous position. From a user's perspective, this will appear as if the instruments are in a different position in the heart (example, at one instance the catheter could be positioned inside the heart, and the next, it will appear as if outside the heart altogether).

At present, the operator's sole recourse is to reset the system, which requires remapping the endocardium and results in lost of previous data (electro-anatomic map, and ablation points). This is highly undesirable.

Preliminary Work

This research is still in progress.

Computer Assisted Mosaic Arthroplasty

Osteoarthritis is the most common form of arthritis and a degenerative joint disease. The cartilage in joints is different than that found in other cartilaginous structures in the human body. Once damaged (due to injury or "wear and tear") it does not naturally recover (at least not quickly, as compared to the progression of the damage). It is generally quite painful, and the common practice (in the longterm) is to perform joint replacement (generally reserved for older patients, as the artificial joint has a limited lifespan).

One alternative is to perform mosaic arthroplasty (also known as, osteochrondal implantation). In this procedure, the orthopaedic surgeon harvests joint cartilage from non-load bearing parts of the joint and implants them at the damage site (ie. the lesion). (This is analogous to filing in a pot hole.)

While traditional mosaic arthroplasty provides an interesting alternative to joint replacement, it is a difficult procedure. The surgeon has to recreate the original joint surface near perfectly. If they insert the replacement plugs to high/proud, they shear off. If they place them to shallow, they fill in with a less desirable cartilage tissue (fibrocartilage versus articular cartilage). If the surgeon leaves too large a gap below the plug, then it can fill with fluid and form a type of cyst.

The use of computer intervention removes much of the human error and guess work out of the procedure. The process used in my (collaborative) work has taken two forms, (1) the use of thermo-plastic custom 3D printed templates (essentially customized jigs), and (2) the use of opto-electronic tracking of surgical tools

Initial Step

The initial step is to acquire the preoperative diagnostic CT (or MR) and segment the joint (identify the joint from the background and surrounding tissue). The segmentation can be achieved semiautomatically using thresholding, edge detection and region growing methods, but the region of interest (ie. the lesion) is normally manually segmented (or edited). Then the patient's ideal curvature is recreated either by consultation with surgeon using a manual planner (designed by Dr. Kunz and others) or using planning software originally designed by Dr. Jiro Inoue PhD. Using the reconstructed joint, surgical sites are planned.

Method 1 - Custom Template Guides

One of the two methods developed for fixing lesions is via thermo-plastic templates. Using the reconstructed and patient's knee models, surgical sites are planned. These plans are then used to design thermoplastic templates that are 3D printed. These templates are design to fit the patient's joint and guide the surgical instrument from the harvest to the recipient site (accounting for 6 degrees of freedom -- illustrated in the above photos).

My work has shown this method to be significantly more accurate (in customized bone models) than the conventional procedure. This method leaves less control in the surgeon's hands, which reduced variance (ie. the outcome from procedure to procedure and surgeon to surgeon is less than in other methods). The shortcoming is that this method requires a more invasive opening of the joint cavity as compared to the opto-electronic method.

Method 2 - Optical Electronic Guidance

The other method developed for repairing lesions involves attaching optical LED markers and tracking the instruments using a stereoscopic camera (This is analogous to tracking done in true-motion or video gaming consoles like the Nintendo Wii, but with significantly greater reliability and accuracy). In this method, the surgical plan is overlayed on the patient's (damaged joint) model. The optical-electronic system is calibrated using 3D printed and machined calibration blocks. Then using an ICP-variant, the patient's joint is reconciled (registered) with the computational model. The navigation system then tracks the instruments relative to the joint surface, providing visual feedback (6 DOF) to the operator/surgeon relating.

This process is significantly more accurate than the conventional procedure (and non-significantly marginally less accurate than the template method). This method is more time consuming due to possible loss of registration and calibration steps, but has the advantage of being usable in less-invasive athroscopic variants of the procedure.

Validation

An important aspect of my masters research was the validation of these techniques. This was performed on composite bone models (sawbones) in which real-patients defects were introduced on the joint surface.

More Detailed Reading

Sebastyan, S., Kunz, M., Stewart, A. J., & Bardana, D. D. (2015). Image-guided techniques
improve accuracy of mosaic arthroplasty. International journal of computer assisted radiology and
surgery, 1-9. LINKED HERE

Sebastyan, S. (2013). Computer-assisted mosaic arthroplasty: A femur model trial. LINKED HERE