Microcalcification localization and cluster detection using unsupervised convolutional autoencoders and structural similarity index

Conference Paper

Detecting microcalcification clusters in mammograms is important to the diagnosis of breast diseases. Previous studies which mainly focused on supervised methods require abundant annotated training data but these data are usually hard to acquire. In this work, we leverage unsupervised convolutional autoencoders and structural similarity (SSIM) based post-processing to detect and localize microcalcification clusters in full-field digital mammograms (FFDMs). Our models were trained by patches extracted from 3,632 normal cases, in total with 16,702 mammograms. Evaluations were conducted in three aspects, including patch-based anomaly detection, pixel-wise microcalcification localization, and microcalcification cluster detection. Specifically, the receiver operating characteristic (ROC) analysis was used for patch-based anomaly detection. Then, a pixel-wise ROC analysis and a cluster-based free-response ROC (FROC) analysis were performed to assess our detection algorithms of individual microcalcifications and microcalcification clusters, respectively. We achieved a pixel-wise AUC of 0.97 as well as a cluster-based sensitivity of 0.62 at 1 false positive per image and 0.75 at 2.5 false positives per image. Both qualitative and quantitative results demonstrated the effectiveness of our method.

Full Text

Duke Authors

Cited Authors

  • Peng, Y; Hou, R; Ren, Y; Grimm, LJ; Marks, JR; Hwang, ES; Lo, JY

Published Date

  • January 1, 2020

Published In

Volume / Issue

  • 11314 /

International Standard Serial Number (ISSN)

  • 1605-7422

International Standard Book Number 13 (ISBN-13)

  • 9781510633957

Digital Object Identifier (DOI)

  • 10.1117/12.2551263

Citation Source

  • Scopus