Abstract Colon cancer organoids hold great potential for studying the development of colon cancer. To assess their physiological and developmental status, both morphological parameters (e.g., cell shape and volume) and functional parameters (e.g., presence of proliferating cells or pluripotent stem cells) are crucial. However, current fluorescence labeling methods are invasive, potentially damaging or altering the organoids, making them disposable. To overcome this limitation, it is essential to develop a non-invasive method for monitoring organoids for long-term. We propose a solution that integrates holotomography and deep learning. Holotomography is a label-free imaging technique that measures the 3D refractive index distribution of samples without requiring any exogenous labeling. While this technique operates within a label-free framework, the resulting images provide only structural information and lack functional information. This limitation makes it challenging for researchers to distinguish cellular and subcellular structures with specific function. We address this challenge by leveraging deep learning to provide both structural and functional information through virtual staining of label-free images. Specifically, we construct a paired dataset using correlative holotomography, which captures both label-free images and their corresponding fluorescence images. Using supervised learning, we train a neural network to map label-free images to their corresponding fluorescence counterparts. Once trained, the neural network can automatically predict virtual fluorescence images from label-free inputs without exposing the organoids to any exogenous chemicals. To validate the accuracy of the trained neural network, we first acquire label-free images of the organoids and input them into the network. After obtaining the network-generated virtual fluorescence images, we chemically stain the same organoids. Finally, we compare the chemically stained images with the virtual images produced by the neural network to evaluate its performance. By combining the safety of label-free imaging with the precision of artificial intelligence, our technology eliminates critical barriers to organoid research and translation. This breakthrough accelerates the development of scalable and safe organoid therapies, bringing us closer to a future where lab-grown organoids revolutionize regenerative medicine and save lives. Citation Format: Juyeon Park, Geon Kim, YongKeun Park. Developing virtual staining algorithm of colon cancer organoids using holotomography and deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7448.
Read full abstract