The research of Joshua D. Trzasko focuses on the creation of mathematical and computational tools for inverse problems in medical imaging, predominantly for magnetic resonance imaging (MRI). Unlike digital cameras, most medical scanners do not measure image data directly. Instead, images are generated from raw data via computer algorithms that leverage a priori knowledge about the physics and measurement uncertainty statistics of the imaging device as well as of the target anatomy or physiology.
The goal of Dr. Trzasko's work is to develop novel and efficient image reconstruction methods that maximize the amount, accuracy and precision of information present in the generated medical images, and thus facilitate increased diagnostic confidence in a wide range of applications, organ systems and diseases.
Fast imaging. For MRI, in particular, creating higher resolution images requires collecting more data. This in turn requires keeping the patient inside of the scanner for more time and limits temporal resolution of physiological processes such as contrast-enhanced blood flow.
Compressive sensing and low-rank matrix methods are modern signal processing concepts that enable high-resolution images to be accurately reconstructed from much less data than traditionally required, by leveraging abstract a priori knowledge about the target image series, such as that cardiac motion is approximately periodic. An outstanding challenge of using this technology is that reconstruction algorithms are computationally expensive, and much of the current research focus in the area is on developing fast computational methods that can execute in clinically feasible times.
Artifact correction. The standard image reconstruction methods employed by commercial MRI scanners generally assume ideal patient and hardware scenarios. In practice, however, there are many nonideal scenarios, such as patient motion, nonstandard hardware or magnetic field perturbations at air-tissue interfaces. These factors cause representation errors in standard reconstruction, and the resulting images contain distorted artifacts that can confound diagnosis.
Dr. Trzasko is working to develop new reconstruction methods that prospectively account for these nonideal scenarios; more artifact-free images can often be recovered from the very same data, avoiding the need for rescanning or switching to another modality.
Quantitative MRI. The numerical values of pixels in standard anatomical images derived from MRI are not inherently quantitative. Thus, most MRIs are qualitatively interpreted based on the morphologies and relative contrasts of structures.
However, if a series of images is collected using MRI, it often is possible — though not trivial — to extract quantitative information such as fat fraction, tissue stiffness, perfusion rates or material relaxation constants based on knowledge about the physics of the acquisition protocol. There are many open problems in these areas, particularly with respect to ensuring that processing techniques consistently provide both accurate and precise results.
- Signal processing theory. Most open problems in MRI reconstruction are not really specific to MRI or even to medical imaging-specific — they are fundamentally the same open mathematical problems that exist in many other disciplines. Much of Dr. Trzasko's work focuses on developing and translating core mathematical techniques in the domain of signal processing theory for solving these problems. One recent example involved the translation of mathematical techniques developed for Netflix's movie recommendation system to obviating the need for patient-specific calibration processes in MRI.
Significance to patient care
Medical imaging, and particularly MRI, is an incredibly powerful diagnostic tool, but also one that is often expensive and uncomfortable for patients. By maximizing the quality and amount of diagnostic information in medical images, Dr. Trzasko helps provide patients with the best value for their time and money. Moreover, by expanding the capabilities of MRI, this technology can potentially be used for diagnosing a wider range of clinical problems.
- Junior Fellow, International Society for Magnetic Resonance in Medicine (ISMRM), 2012
- First Place Poster Award, ISMRM, 2011