Specialty Area: Small Animal Imaging

Jim Bankson jbankson@mdanderson.org

Highlights

• Overview of the data processing pipeline

• Survey of common measurements from MRI

• Software platforms for data analysis

• Validation and quality assurance

• Data organization and management

• Summarizing complex multi-dimensional data

Data Analysis

Magnetic resonance imaging is a powerful tool in the biomedical sciences because it produces images with exquisite soft-tissue contrast and relatively high spatial resolution. Imaging protocols can be designed to provide a wide range of information about the structure, function, and composition of tissues in vivo. In addition to an understanding of the underlying biological question and of fundamental principles of MRI that scales with the complexity of the measurements, the analysis of complex, multi-dimensional MRI data requires a suite of tools and processes that should work together to provide a robust and efficient pathway from data acquisition to distribution and dissemination. Some of these tools may be present on the MRI console itself, while other kinds of analyses may require the use of 3rd party software or implementation of new purpose-built scripts. In this lecture, we will discuss the data processing pipeline for experimental MRI and survey some widely available tools that can facilitate the process. We will also discuss strategies that can be employed to enhance the reproducibility of research results. The

Data Processing Pipeline

The first stage of the process involves image reconstruction, which is generally automatically performed by the MRI console but may be performed offline for unconventional encoding strategies or to eliminate scaling, filtering, or other automated manipulations for image display. One or more images may be analyzed together; some analyses can be made directly on the scanner, while other measurements must be made using 3rd party software or analyzed using custom-made algorithms. The output of this stage may range from a single number (length, or volume for example) to complex multi-dimensional parametric data. Cohorts of these measurements are then needed to test for experimental differences between groups. Finally, all of this data must be distilled into a descriptive summary of the experiment and the critical information that has been learned.

Representative Measurements

Simple and Complex MRI excels at supporting straightforward measurements to assess characteristics, such as the size or shape of a tumor, lesion, or other target anatomy, which may be difficult or impossible using other imaging modalities. A wide variety of contrast mechanisms can be exploited in order to learn much more about the structure and function of tissue and disease, from quantitative relaxometry to measurements of flow, velocity, diffusion, perfusion, and even metabolism through spectroscopic imaging of hyperpolarized substrates.

Software Platforms

Basic measurements of size, shape, and signal intensity can typically be made directly on the scanner. More advanced analysis of relatively standardized measurements may also be available as optional software features. Imaging data is often exported for analysis using a variety of open source or commercially available software. Scientific scripting languages enable tailored analysis and the development and testing of new methods for image reconstruction and data analysis. If this approach is used, it is very important to implement a version control system to ensure that an analyses can be easily repeated, perhaps years later, despite the use of algorithms that may evolve over time.

Validation & Quality Assurance

Once the process for making these measurements has been established, it is very important to validate measurements against known imaging phantoms. A reliable quality assurance program gives confidence that scanner performance remains consistent. Measurements derived from imaging phantoms with known characteristics are very useful for training purposes and validation of any changes in the imaging sequence or analysis algorithm.

Data Organization & Management

A little time spent planning for a logical way to store and organize data can pay dividends, particularly when the need arises to revisit data at a later time when the experiment is no longer in recent memory. Clear records, organization, and automation of as much of the process as possible minimizes the likelihood of data entry errors that can prove disastrous in the final analysis. When a large cohort of data is acquired as part of a collaborative research project, consider the accessibility of the data (format and structure) for collaborators that might access to imaging data and analyses.

Summarizing Complex Multi-Dimensional Data

Imaging experiments produce a very large amount of data that can be overwhelming and turn otherwise important results into a proverbial needle in the haystack. Consider the interests and expertise of the target audience; some will prefer representative images and summary statistics that clearly and succinctly summarize results, while others may prefer to explore details.

Discussion

MRI is a complex and flexible imaging modality that supports a wide range of biomedical measurements and research. There are many components to data analysis, and a wide range of approaches to the end result. Fortunately, there are a variety of tools available to help support efficient and reproducible analysis of preclinical imaging data.