|
Abstract:
|
The detection of myeloid cells is an important oncology research area because their presence is correlated with the patient’s prognosis and response to therapies. However, performing this task on H&E is challenging and impractically time-consuming even for a trained pathologist. For this reason, immunohistochemistry or immunofluorescence using antibodies against myeloid specific markers are used for their identification. However, these methods require additional tissue samples and expenses. Therefore, there is a need to have a tissue-preserving and low-cost method to detect myeloid cells on H&E sections. In this study, we propose a deep learning myeloid cell detection model that takes a H&E tissue and returns a prediction of the myeloid cells. First, we developed a deep learning model to extract the IHC stained region from a IHC image, which is used as ground truth labels for training a myeloid region segmentation model. Next, we train a cell classification model to classify the nucleus of the myeloid cells from the detected cell nuclei inside the region of interest. As proof of concept, we will present results from a macrophage cell detection model.
|
