Skip to content

Advertisement

  • Poster presentation
  • Open Access

A novel threshold-independent computer-aided detection algorithm for breast MRI

  • 1,
  • 2,
  • 1 and
  • 1
Breast Cancer Research201012 (Suppl 3) :P15

https://doi.org/10.1186/bcr2668

  • Published:

Keywords

  • Motion Artefact
  • Reduction Algorithm
  • Tumour Detection
  • Manual Analysis
  • Benign Tissue

Introduction

Image degradation due to motion artefact in breast MRI represents a diagnostic challenge. Tumours are often detected manually by a radiologist or with computer-aided detection (CAD) systems, which utilise areas of enhancement that meet a predefined threshold. The aim of this study was to test a new threshold-independent CAD algorithm and to correlate its findings to the conventional manual analysis.

Methods

CAD was tested on retrospectively acquired MRIs of 14 patients with pathologically proven carcinomas. CAD results were obtained in a fully automated manner and the expert was blinded to the CAD findings. Noise artefacts were eliminated with the patient motion reduction algorithm and suspicious tissues were delineated using a novel all-timepoint-based, threshold-independent parametric map approach. The algorithm evaluates the shape of the curve as a whole and uses the noise integral to the image to discriminate malignant from benign tissues.

Results

All CAD-identified tumours and generated kinetic curves were comparable with those of the manual analysis. In particular, tumour conspicuity was enhanced in two cases where image degradation by motion artefacts made data interpretation challenging to conventional analysis. See Figure 1.
Figure 1
Figure 1

Left: tumour visualisation with conventional analysis DCE-MRI. Middle: CAD curve shape map (red, wash-out; green, plateau; blue, persistent). Right: kinetic curve.

Conclusions

CAD results were favourably viewed by experts and 100% correlated to conventional manual tumour detection. In particular, CAD appears to increase tumour conspicuity in cases with motion artefacts. Prospective analysis is required to test this model further.

Authors’ Affiliations

(1)
Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield, UK
(2)
Image Analysis Ltd, Leeds, UK

Copyright

Advertisement