Special Series Guest Editorial: Artificial Intelligence and Machine Learning in Biomedical Optics

Abstract

Artificial Intelligence (AI) methods, including machine learning (ML) and deep learning (DL), are quickly evolving, and impacting a very wide range of scientific endeavors. Biomedical optics is no exception and AI methods are currently transforming our discipline on an almost daily basis. From changing data acquisition 1,2 and image reconstruction methods, 3 to segmentation and interpretation of optical images, 4 AI methods are providing improved solutions to established problems and enabling new problems to be addressed. Structured light can be combined with AI methods to probe and interpret the interaction of light with biological tissues. For example, the coupling of AI methods with hyperspectral and multispectral systems can enable the detection of specific molecular signatures in tissue, cells, and biofluids. 5,6 Supervised ML/DL methods are well-suited for this purpose, since they can implicitly learn high-dimensional image statistics and complicated mappings that describe optimal decision strategies for a variety of inferences of relevance to basic science and clinical applications. Enhancing advanced optical methods with AI will enable the clinical translation of new optical sensing and imaging technologies. Label free optical imaging, such as stimulated Raman histology, hyperspectral imaging, and convolutional neural networks (CNNs), has been successfully employed for intraoperative automated brain tumor diagnosis with near real-time detection. 7,8 Integrating ML/DL methods with optical methods such as coherent anti-Stokes Raman scattering imaging, optical colonoscopy and fluorescence lifetime imaging has shown to be effective in the differential diagnosis of lung cancer, 9 colorectal cancer, 10 and cervical neoplasia, 11 respectively. Another AI-enabled game-changer will be the use of DL methods for computational staining of label-free optical images, resulting in all-digital histopathology. 12-14 In clinical decision making, where accuracy and timing can be critical, spatial frequency domain imaging coupled with ML has been employed for predicting the severity of burn injuries. 15 The combination of multi-photon imaging with ML/DL has further enabled improved lymphedema diagnosis, 16 skin cancer screening 17 and atopic dermatitis. 18 ML combined with emerging feature engineering approaches has become the mainstay in tissue, cells, and biofluids interrogation in spectroscopic methods. Examples of such applications range from neurosurgical guidance using spontaneous Raman spectroscopy for cancer detection 19 to detection of aggressive variants of prostate cancer in pathology using Raman micro-spectroscopy. 20 Merging optical coherence tomography (OCT) imaging with AI provides a unique opportunity to analyze this plethora of information and assist in making clinical decisions in the field of ophthalmology with applications in retinal imaging, 21 glaucoma 22 and age-related macular degeneration. 23 *Address correspondence to Behrouz Shabestari, behrouz.shabestari@nih.gov