Knowledge Transfer across Breast Cancer Screening Modalities: A Pilot Study Using an Information-Theoretic CADe System for Mass Detection

  • Authors:

  • Georgia D. Tourassi;Amy C. Sharma;Swatee Singh;Robert S. Saunders;Joseph Y. Lo;Ehsan Samei;Brian P. Harrawood

  • Affiliations:

  • Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710;Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710;Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710;Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710;Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710;Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710;Duke Advanced Imaging Laboratories, Department of Radiology, Duke University Medical Center, Durham, USA NC 27710

  • Venue:
  • IWDM '08 Proceedings of the 9th international workshop on Digital Mammography
  • Year:
  • 2008

Quantified Score

Hi-index 0.01

Visualization

Abstract

We have performed a series of experiments to assess whether a featureless, knowledge-based CADe system that relies on information theoretic principles is capable of transferring knowledge across cases acquired with different imaging modalities. Typical feature-based CADe systems are developed and carefully optimized for a specific imaging modality and platform, namely for screen-film mammograms (SFMs) digitized with a specific digitizer, or for full-field digital mammograms (FFDMs), or for the newly introduced digital breast tomosynthesis (DBT) modality. Multiplatform application of such CADe systems is often limited due to image processing steps that are tailored to the imaging modality and acquisition protocol. It is desirable however to develop CADe systems with the ability to adapt to a dynamically changing environment (i.e., imaging modality) and provide an accurate decision while capitalizing on knowledge acquired at different, yet related environments. Working towards this goal, we present a pilot study using a knowledge-based CADe system for mass detection that uses information theory to assess the similarity between a query and a knowledge case. We evaluate the system's ability to transfer knowledge across three imaging modalities (SFMs digitized with two different digitizers, FFDMs, and DBTs) when performing the detection task. Overall our study showed that although blind translation of the system in a new modality for which no prior knowledge exists results in reduced performance, the system is still able to operate at a very decent level. When the system operated with a knowledge database of mixed cases, its performance was robust and comparable to what observed independently.