Graduate Research Assistant - UT MD Anderson Cancer Center - Dr. Moiz Ahmad

The University of Texas M. D. Anderson Cancer Center

Houston, TX

ID: 7101995
Posted: June 12, 2022
Application Deadline: Open Until Filled

Job Description

We are looking for a graduate research assistant in the area of medical image analysis to join a newly established research lab. Quantitative imaging research is a key component to enabling and guiding personalized oncological patient care. The Tumor Measurement Initiative (TMI) STRIDE aims to build an institutional platform to support standardized, automated, quantitative imaging-based tumor measurement across each patient’s journey to advance multidisciplinary, data-driven, high-precision cancer treatment. Specifically, to process and analyze imaging data, including image quality, measurement tools, and newly developed automated segmentation tools (for tool evaluation and validation).

Learning Objectives:
The research topics are based on: 1) imaging protocol standardization 2) development of clinical image-based quality metrics 3) correlation of quality metrics to accuracy of automated image analyses, including segmentation, classification, and prediction. Specific project descriptions are provided below.

Project for the position:
Project 1: Protocol standardization and quality assurance
Clinical volumetric imaging, such as magnetic resonance imaging (MRI) or computed tomography (CT) generally requires complex protocols to acquire high-quality images. The imaging protocol used has a wide-ranging impact on the diagnostic quality of a radiologic study impacting image contrast, vulnerability to artifacts, longitudinal tracking, etc. Therefore, it is critical that the imaging protocols delivered by a radiology practice be consistent and well-controlled. Protocols are generally anatomical site or disease-specific and can vary from vendor to vendor to account for different technical capabilities. Maintaining and monitoring these protocols is challenging especially for a large distributed radiology practice like MD Anderson. Fortunately, most protocol information is stored in the image header and is ripe for data mining and aggregation which could greatly assist in protocol management. Commercial tools for protocol management are nascent and tend to lack the multi-vendor compatibility or flexibility needed for MD Anderson’s practice. We propose to use an imaging research platform (XNAT) to implement protocol quality assurance tools that will allow for delivered protocol auditing which will assist in the standardization and improvement of both the clinical and research protocols performed at MD Anderson. The metadata of clinical imaging data stored on XNAT shall be analyzed to determine consistency of imaging protocols across imaging systems and comparable imaging protocols. Year 1 tasks will include but not be limited to:
- Analysis of metadata for clustering and outlier detection for determination of needed clinical changes
- Defining comparable imaging exams, mining the data, and determining what percentage of exams have been performed according to protocol:
o Radiation dose compliance to protocol
o Acquisition and image reconstruction parameter consistency (i.e., what percentage of exams used the correct and/or expected parameters per the protocol?)

Project 2: Clinical Image-Based QA Metrics
Given the critical nature of image quality for radiologic diagnostic performance, robust systems for quality assurance have been developed. Generally, these systems are based on phantom studies to assess equipment performance and safety. While useful, these methods do not test the system response to all clinical conditions and must be physically acquired limiting the frequency they can be performed. The clinical images themselves are a rich source of quality information as they are the ultimate delivered clinical product and are also acquired with high frequency. However, objective, accurate, automated assessment of clinical images remains a nascent research area. The volume of images performed at MD Anderson represents an ideal data set to evaluate clinical image-derived quality metrics and how they can be leveraged to improve practice.
Algorithms for quantifying image signal-to-noise, image contrast, exogenous contrast enhancement, and image artifact identification shall be developed and validated. Validated algorithms will be used to assess equipment performance as well as clinical protocol performance. Year 1 tasks would include but not be limited to:
- Applying automated algorithms that measure imaging quality metrics such as CT global noise and MRQY to clinical exams and creating image quality metric datasets separated by protocol or diagnosis.
- Validate existing algorithms (MRQY and CT Global Noise) for clinical application across different body sites and imaging procedures.
- Correlate consistency in Project 1 and determine association with consistency of imaging-based QA metrics
Project 3: Correlation of image quality to accuracy of disease detection and classification system
Radiologic images likely contain vastly more information than is currently being used by human interpreters for patient care. Imaging features have begun to show promise for their ability to tap this potential well of information, especially when using radiomics and machine learning tools. The use of image quality metrics may improve the performance of these tools or otherwise provide an indicator of diagnostic confidence in these tools. Radiomics or machine-learning insights backed by image quality and confidence scores could greatly augment clinical imaging data and improve the quality of clinical information provided to clinicians in MD Anderson’s radiologic practice.
The relationship between image quality and tumor classification algorithms shall be assessed. The accuracy of existing algorithms (e.g. radiomics models) will be re-evaluated with image quality score included as an extra parameter. Additionally, prediction models shall be developed that use image quality scores to predict the accuracy of radiomics models. Year 1 and 2 tasks would include correlation related objectives such as:
- Building on outcomes of projects 1 and 2 to determine impact on radiomic predictive capabilities and auto-segmentation projects. (e.g., impact of image quality on malignancy prediction and NVIDIA brain tumor segmentation)
- Collaboration with Siemens Alliance for Standardizing Imaging Acquisition Protocols and evaluating outcomes of PQI with imaging-based quality metrics that could lead to automated assessment of clinical image quality
- Combining image quality metrics with images that are with and without artifacts in order to develop algorithms for automatic detection of artifacts. auto-detect artifacts
Additional Future Tasks (timing TBD):
- Research and development of additional algorithms such as automated ways to measure blood vessel contrast
- To analyze and redesign localizers for optimization using imaging-based QA metrics to maximize capture of quality data for every patient, every scan

Eligibility Requirements:
Candidates from a diverse background are encouraged to apply. The successful applicant must be enrolled in a Ph.D. program in the fields of Biomedical Engineering, Medical Physics, Applied Mathematics, Computer Science, Electronic Engineering, Bioinformatics, or related fields.
Preference is given to candidates with demonstrated experience in data science or machine learning, and proficiency in either MATLAB or Python programming languages.

Dates or Training Schedule:

Mentor Matching:
This position reports to (Proposed joint supervision: Dr. Ahmad, Dr. Walker, Dr. Yung)

Interested Candidates, please email your CV to Dr. Moiz Ahmad at [email protected] or Jeannette McGee at [email protected]

It is the policy of The University of Texas MD Anderson Cancer Center to provide equal employment opportunity without regard to race, color, religion, age, national origin, sex, gender, sexual orientation, gender identity/expression, disability, protected veteran status, genetic information, or any other basis protected by institutional policy or by federal, state or local laws unless such distinction is required by law.