Summary
With the integration of 3D cellular-resolution optical coherence tomography (OCT) and deep learning networks, we will segment the nuclei, epidermis/dermis junction, tissue pigment, microcirculation, inflammatory cells, and even the border of a malignant tumor. The epithelial tissue manifest of chronic diseases will be studied for early clinical disease diagnosis and risk assessment.
Advantages
The acquisition of OCT images is based on the 3D high-resolution tomography imaging system made by the Apollo Medical Optics, who received technology transfer from our team. It would be helpful for clinicians if the OCT images could be translated to the H&E stained images. Physicians' medical knowledge will be incorporated to train the deep learning algorithm for accurate image translation. This project will introduce self-supervised learning and various annotation methods (e.g., physician annotation, H&E/IHC image color separation, dynamic OCT analysis algorithm). In addition to the nuclei distribution and epidermis anatomy, the converted stained images will provide quantitative information on pigment distribution, microcirculation, and inflammatory cells. Both un-supervise and supervised algorithms will be employed. The labeling of inflammatory cells will be obtained by immunofluorescence staining.
Sub-project 7 will build an image database to explore the manifestation of the skin or epithelial tissue in early-stage chronic diseases through clinical trials. We will continue to work with Memorial Sloan Kettering Cancer Center in New York to detect and segment human OCT images of skin cancers to assist a rapid clinical diagnosis.
Starting from the segmentation of nuclei and epidermis/dermis junction, we will also conduct the annotation and segmentation of tissue pigments (such as melanin, hemoglobin, etc.), microcirculation, inflammatory cells, and malignant tumor boundaries. In-depth study of the epithelial tissue manifestation of chronic diseases and upgrading OCT cell-level image quality are the two focuses of the project. Eventually, using deep learning algorithms from morphology changes to tissue functions to assist physicians in the early diagnosis of chronic diseases.
Applications
- Clinical diagnosis of skin and epithelial diseases
- Medical image translation between OCT and stained images (e.g. H&E, IHC)
Keywords
Optical coherence tomography, deep learning, image translation, tumor boundary segmentation
◎ PI
