BROWSE
Detail
Conference Paper
An Automated Breast Mass Detection Algorithm on Digital Breast Tomosynthesis Images using Hough transform and Convolutional Neural Networks
Cited - time in 
Share 
Share
- Authors
- Ji-Wook Jeong, Seung-Hoon Chae, Sooyeul Lee, Eun Young Chae, Hak Hee Kim, Young Wook Choi
- Issue Date
- 2017-01
- Citation
- International Forum on Medical Imaging in Asia (IFMIA) 2017, pp.166-168
- Language
- English
- Type
- Conference Paper
- Abstract
- We suggest and test a simplified computer-aided detection (CADe) system for breast masses in reconstructed digital breast tomosynthesis (DBT) volumes using a three-dimensional (3D) Hough transform and convolutional neural networks (CNN). The system consisted of masking, subsampling, a contrast adjustment, a 3D Hough transform, and a CNN-based false-positive (FP) reduction steps. A 3D Hough transform can easily generate a list of Hough spheres around a breast mass in order of their votes, and the overlapped spheres are in principle combined into be a larger sphere covering the effective range of the examined breast mass candidate. We obtained the FP rate of 4.84 per DBT volume with a sensitivity of 92.8% using A 3D Hough transform-based mass candidate detection steps, and the further enhanced FP rate of 3.31 with a sensitivity of 90% by applying the FP reduction step involving a typical CNN-based training model.
- KSP Keywords
- 3D Hough transform, Breast mass detection, Computer-aided Detection(CADe), Contrast adjustment, Convolution neural network(CNN), Detection algorithm, Digital breast tomosynthesis(DBT), Effective range, FP reduction, Three dimensional(3D), training model