Development of new breast X-ray imaging technologies or improvements to hardware or software of current systems usually require the accurate assessment of image quality. Image quality assessment methods are also required for quality control (QC) of clinical systems, for example as required by the U.S. Mammography Quality Standards Act (MQSA) program. The gold standard for assessment of image quality is human reader studies assessing diagnostic performance over a cohort of representative clinical images. These clinical trials are often difficult and expensive to perform, and therefore researchers have been studying alternative approaches that can assess diagnostic task performance without imaging patients.
This short course will discuss methods for objectively assessing task performance of breast imaging systems without conducting a clinical trial. One approach that will be discussed is the in silico modeling of a clinical trial. This approach involves complete computer modeling of each step in the imaging chain including: 1) modeling of breast and relevant breast lesions, 2) modeling of the imaging system, and 3) modeling of the observer. Another more experimental approach that will also be discussed involves: 1) development of anthropomorphic physical phantoms with diagnostic features, 2) imaging of these phantoms on breast imaging commercial or prototype systems, and 3) assessment of task performance with either model or human observers.
For maximum efficiency, the proposed in silico and experimental approaches require the development of computer or model observers that can emulate either ideal or human observer task performance. This short course will discuss the use of new machine learning algorithms that can be used to model observer performance in the assessment of breast imaging technology.
This course will describe and make attendees aware of useful open-source software tools that can be downloaded.
- Summarize new methods for generating both digital and physical anthropomorphic breast phantoms that can be used for objective technology assessment.
- Identify new open-source software that can be used to generate digital breast phantoms.
- Summarize an approach for in silico Monte Carlo modeling of breast imaging systems.
- Describe and learn about approaches to validate modeling tools.
- Identify open-source Monte Carlo software available for in silico modeling of breast imaging systems.
- Describe various approaches to modeling observers for breast phantom images. These include both conventional and machine-learning based model observers.
- List advantages and disadvantages of various model observers.
- Describe a new approach for objectively assessing breast imaging detectors.
- Explain how to use machine learning model observers to assess task performance achieved with various breast imaging modalities.
- Describe and learn about an in silico clinical trial conducted by the FDA to compare task performance with full-field digital mammography and digital breast tomosynthesis.