DIB technology is an imaging biomarker derived from large-scale medical image data. Letting the machine define important diagnostic features by itself without guidance from previously established medical criteria (defined by humans) is key to our deep learning technology. By expanding beyond the status quo scope of diagnostic criteria, we envision a constructive partnership between physicians and our technology in becoming "better together" faced with challenges in accurate diagnoses of diseases.
Invented in 1900, chest radiography has been, and still is, the single most commonly performed diagnostic imaging test. It is mainly used to evaluate the lungs and heart, help diagnose and monitor treatment response for a variety of diseases such as pneumonia, tuberculosis and lung cancer. Albeit an old technology, it remains to take an integral part of clinical decisions made by physicians worldwide.
Despite the large amount of collective experience in interpreting chest x-rays, a significant number of cases are still misinterpreted and misdiagnosed on chest x-rays. Even though chest radiography is a basic part of medicine, it remains to be a challenging task to accurately interpret chest x-rays. There is room for improvement in terms of accuracy and consistency both due to the inherent limitations of the modality itself as well as limitations of the human visual system.
Our research aims to use technology to understand lesions on chest x-rays in depth and devise better models of lesion morphology in order to improve the overall diagnostic performance of chest radiography interpretation.
First developed in 1965, mammography has been the primary imaging modality for breast cancer screening, proved to have survival benefit. Implemented in national cancer screening programs throughout the world, it seeks to provide earlier detection of breast cancer, and is also used to assess extent of disease and treatment response.
Despite technological advances to enhance mammography, there many false-negative cases especially in subjects with dense breast. A significant rate of false-positive cases associated with mammography is also problematic as there are many women are subjected to unnecessary painful invasive procedures due to the erroneous result on the mammography, not to mention the psychological distress they have to experience. Mammography in its current form is far from perfect in its accuracy. Due to its common use worldwide, even small improvements would benefit a vast number of women.
Our research aims to apply advanced algorithms to analyze mammography images in fine detail and develop improved detection models for malignant features in order to significantly decrease false negative and false positive results.
Whereas chest radiography and mammography is integral in the initial detection of disease, pathology is the pillar of the final diagnosis process in medicine. Visualized in its most basic unit, the cell, within its community, the tissue, diseases are assessed in conclusion on their identification and state.
Despite its importance in the diagnostic process, there is a great deal of inconsistency involved in the process. Lack of quantifiable objective standards and the tedious nature of the interpretation process both attribute to the inconsistency. Much improvement in the diagnostic process by application of analytics is feasible; moreover, with the recent advent of digital pathology, the applicability of analytics has been greatly enhanced.
The purpose of our research in digital pathology is to objectively define the myriad morphological features in tissue samples and innovate the accuracy, efficiency, and consistency of histopathological diagnosis.