Widened 3D Field of View Enhances OCT Diagnostic Imaging
Researchers at Duke University developed an optical coherence tomography (OCT) technique that delivers high contrast and high resolution over a wide, 3D field of view. The enhanced OCT technique, called 3D optical coherence refraction tomography (3D OCRT), produces highly detailed images that reveal features difficult to observe with traditional OCT.
By providing clearer imaging and enhanced resolution, the approach to OCT could improve medical diagnostic imaging. According to researcher Kevin Zhou, the approach takes the existing functionalities of the volumetric imaging technique and adds an extension. The extension features novel hardware combined with a computational 3D reconstruction algorithm. This combination aims to address well-known limitations of OCT, he said. Enhanced OCT images of a fruit fly head. The method produces highly detailed images that reveal features difficult to observe with traditional OCT. Courtesy of Kevin Zhou/Duke University. The researchers created an optomechanical design that features a parabolic mirror as the imaging objective. The design makes it possible to image a sample from multiple views over a wide range of angles, allowing the researchers to acquire OCT volumes from multiple views over a range of up to ±75° without rotating the sample. Although parabolic mirrors are known to exhibit “perfect” focusing only when the incident beam is parallel to the mirror’s optic axis, and therefore are rarely used as imaging objectives, the researchers demonstrated millimetric field of views using the weakly focused, long-depth-of-field beams preferred in OCT. To combine the views into a single, high-quality 3D image corrected for distortions, noise, and other imperfections, the researchers developed a 3D reconstruction algorithm that leverages differentiable programming frameworks and optimization techniques for solving inverse optimization and image registration problems. The algorithm allowed the researchers to perform dense 3D reconstruction from multiangle OCT volumes across large 5D data sets — in the researchers’ case, about 90 GB in size — using just one memory-limited graphics processing unit. “Because our system generates tens to hundreds of gigabytes of data, we had to develop a new algorithm based on modern computational tools that have recently matured within the machine learning community,” said professor Sina Farsiu, who co-led the research. The researchers demonstrated that 3D OCRT can reveal 3D features that are undetectable by conventional OCT techniques in fruit fly, zebra fish, and mouse samples. Using mouse tissue samples of the trachea and esophagus, they showed 3D OCRT’s potential for medical diagnostic imaging. With traditional OCT, it is difficult to acquire high-resolution images over a wide field of view in all directions simultaneously. OCT images also contain high levels of speckle noise that can obscure biomedically important details. Zebra fish larva. The enhanced version of OCT can image biomedical samples at higher contrast and resolution over a wider 3D field of view than was previously possible. Courtesy of Kevin Zhou/Duke University. 3D OCRT offers the coherent detection sensitivity advantages of OCT. In addition, it offers a speckle-free, incoherent contrast mechanism similar to that of incoherent microscopy, together with multifold enhanced lateral resolution over an extended 3D field of view. “In addition to reducing noise artifacts and correcting for sample-induced distortions, OCRT is inherently capable of computationally creating contrast from tissue properties that are less visible in traditional OCT,” Zhou said. “For example, we show that it is sensitive to oriented structures such as fiber-like tissue.” With a few straightforward enhancements, 3D OCRT could become widely used in biomedical imaging and could enable more accurate medical diagnostic imaging, the researchers said. The team is investigating how to shrink the system and make it faster for live imaging by incorporating new techniques to increase the speed of OCT systems and advancements in deep learning that could improve the speed of data processing. “We envision this approach being applied in a wide variety of biomedical imaging applications, such as in vivo diagnostic imaging of the human eye or skin,” said professor Joseph A. Izatt, who co-led the research. “The hardware we designed to perform the technique can also be readily miniaturized into small probes or endoscopes to access the gastrointestinal tract and other parts of the body.” The research was published in Optica (www.doi.org/10.1364/optica.454860).