- Keywords: Multimodal Deep Learning, Breast Cancer, Molecular Subtypes, Convolutional Neural Network, Attention Mechanism, Mammography, Ultrasound
- TL;DR: To predict the molecular subtypes of breast cancer using multimodal deep learning with attention mechanism.
- Abstract: Accurately determining the molecular subtype of breast cancer is an important factor for the prognosis of breast cancer patients, and can guide treatment selection. In this study, we report a multimodal deep learning with attention mechanism (MDLA) for predicting the molecular subtypes of breast cancer from mammography and ultrasound images. Incorporation of the attention mechanism improved diagnostic performance for predicting 4-class molecular subtypes with Matthews correlation coefficient (MCC) of 0.794. The MDLA can also discriminate between Luminal disease and non-luminal disease with areas under the receiver operating characteristic curve (AUC) of 0.855. This work thus provides a noninvasive imaging biomarker to predict the molecular subtypes of breast cancer.
- Paper Type: both
- Primary Subject Area: Detection and Diagnosis
- Secondary Subject Area: Application: Radiology
- Paper Status: original work, not submitted yet
- Source Code Url: The source code we developed will be publicly available on "https://github.com/Tianyu-Zhang0319/Multimodal_Deep_Learning". If you have any questions, please contact the author (email@example.com).
- Data Set Url: The data used in this study is not a public data set, but comes from the Netherlands Cancer Institute. So it is not convenient to make it public. We will publicly provide some data examples on "https://github.com/Tianyu-Zhang0319/Multimodal_Deep_Learning". If you have any questions, please contact the author (firstname.lastname@example.org).
- 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.
- Source Latex: zip