Abstract PO-080: Deep learning-based analysis of heterogeneity of breast cancer cells using lens-free digital in-line holography

Abstract

Lens-free digital in-line holography (LDIH) produces cellular diffraction patterns from a large field of view that lens-based microscopes cannot achieve. It is a promising diagnostic tool allowing comprehensive cellular analysis with high-throughput capability. Since these diffraction images are far more complicated to discern, conventionally computational algorithms are used to reconstruct cellular images for the diffraction patterns. However, it is inefficient and prone to errors. Here, we developed a deep learning architecture, HoloNet to directly analyze the diffraction patterns from LDIH. In addition to the standard CNN (Convolutional Neural Network), the HoloNet includes a holo-branch that extracts large features from holograms and integrate them with the small features from the standard CNN. Using HoloNet, we accurately predicted the intensity values of ER/PR and HER2 in breast cancer cells and classified four known subtypes of breast cancer cells. Moreover, we applied a HoloNet dual embedding, which learns high-level diffraction features related to breast cancer cell types and the intensities of ER/PR and HER2. This allowed us to identify potential subtypes of breast cancer cells from the cell lines and the fine needle aspiration samples from breast cancer patients. We demonstrate that our HoloNet can enable LDIH to perform more detailed and rapid analyses of subtypes of breast cancer cells. Citation Format: Tzu-Hsi Song, Mengzhi Cao, Jouha Min, Hyungsoon Im, Hakho Lee, Kwonmoo Lee. Deep learning-based analysis of heterogeneity of breast cancer cells using lens-free digital in-line holography [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-080.