Fully-Automated Multi-View Classification for Lesion Interpretation in MammographyDownload PDF

Published: 16 May 2023, Last Modified: 16 May 2023Submitted to MIDL 2021Readers: Everyone
Keywords: Deep learning, Medical Imaging, Multi-View Models, Convolutional Neural Networks, Transfer Learning, Image Segmentation
Abstract: Given the extreme importance in early detection of breast cancer, a compelling search for Computer-Aided Detection (CAD) techniques drove Deep Learning (DL) researchers to investigate potential mammography screening applications. This work proposes the use of sophisticated and recently proposed Convolutional Neural Networks (CNNs) for classification and segmentation of mass and Micro-Calcification (MC) lesions on the INbreast dataset via a novel fully-automated pipeline. For segmentation, an Attention Dense U-Net model is used to provide segmentations for masses for MCs. The classification stage is performed via a Dense Multi-View model, receiving an enriched input with the previous predicted segmentations, achieving performance on par with state of the art for fully-automated classification of breast screening exams (Normal, Benign, Malignant), achieving a 3-Class Mean AUC of (0.79±0.06).
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: both
Primary Subject Area: Detection and Diagnosis
Secondary Subject Area: Application: Radiology
TL;DR: We describe a novel fully-automated pipeline comprised of sophisticated holistic Deep Learning models to segment and classify potential Mass and Micro-Calcification lesions in mammography, achieving results competitive with the state of the art.
6 Replies

Loading