Development of Optical Devices for Digital Medicine

다국어 초록 (Multilingual Abstract)

Advances of technology have made a revolution that interconnects industrial devices and fuses the boundaries of digital, physical and biological spaces. These technologies such as cloud computing, 3D printing technology, big data, internet of things (IOT), artificial intelligence (AI), and maturity of system integrations have been improved every year, changing our daily life quickly in intelligent and convenient ways. In this days, these explosions of technology, changing the way we live and think, is referred to 4th industrial revolution. As we know, every industry is affected by the new waves of technologies, digitalization and connectivity, and the biomedical or medical field is no exception. Healthcare fields have benefited mostly from recent technical improvements, revolutionizing the medical systems in many terms in cost-effective ways. Particularly, “digital medicine” has been recently came into the limelight as one of the uprising fields. In digital medicine, traditional medical devices and diagnostic programs have become miniaturized, digitalized, and automated. As taking advantages of digital medicine, specific fields related to digital pathology, point-of-care (POC) diagnostics, and application of deep learning or machine learning technologies have shown the great potentials not only in biomedical academia but also in the revenues of their markets. It allows to connect devices, hospital equipment, and to accelerate efficiencies in health service such as diagnosis, and to reduce the cost of services. Moreover, interconnection between advanced technologies has been improved the access of healthcare to the places where hospital or medical services are limited. Furthermore, artificial intelligence has shown promising results related to disease screening especially using medical images.

Although fields in digital medicine are prospering, still there are limitations that needs to be overcome in order to provide further advanced health services to patients in the various situations. In digital pathology, improvements of microscopic technologies, internets, and storage capabilities have reduced the time-consuming processes. The simple transformation of microscopic image to digital have successfully alternated many limitations in the analogue histopathology workflow to efficient and cost saving ways. However, tissue staining is currently referred as one of the bottleneck that makes workflow still lengthy, labor-intensive, and costly. In the POC diagnostic fields, various digitalized portable smartphone-based diagnostic devices have been introduced as alternatives to conventional medical services. These devices have provided the quality assurance of diagnostics by taking advantages of sharing, and quantitative analysis of digital information. However, most of these works have been focused on replacing diagnostic process which mostly done in laboratory settings. As medical imaging devices and trained clinicians or practitioners are limited, there are also high demands on clinical imaging-based diagnostics in developing countries.

In this thesis, computational microscope using patterned NIR illumination was developed for label-free quantitative differential phase tissue imaging to bypass the staining process of the pathology workflow. This system overcame the limitations found in the conventional quantitative differential phase contrast in a LED array microscope, allowing to captured light scattering and absorbing specimen while maintaining weak object approximation. Moreover, portable endoscope system was developed integrating the additive production technologies (3D printing), ICT, and optics for POC diagnostics. This innovative POC endoscope demonstrated comparable imaging capability to that of commercialized clinical endoscope system. Furthermore, deep learning and machine learning models have been trained and applied to each devices, respectively. Generative adversarial network (GAN) was applied to our NIR-based QPI system to virtually stain the label-free QPI which look comparable to image that is captured from bright field microscope using labeled tissue. Lastly, POC automated cervical cancer screening system was developed utilizing smartphone-based endoscope system as well as training the machine learning algorithm. 3-5% of acetic acid was applied to the suspicious lesion and its reaction was captured before and after application using smartphone endoscope. This screening system enables to extract the features of cancers and informs the possibility of cancer from endoscopic images.