Applications, Devices, and Methods for Smartphone-Based Medical Imaging Systems

Abstract

As cancer rates continue to increase, new adjunctive tools are needed to augment the skills of clinicians to enable earlier detection and diagnosis, a key to reducing morbidity, mortality, and overall healthcare costs. Autofluorescence imaging (AFI) and multispectral imaging (MSI) systems have the potential to increase detection rates in oral cancer and skin cancer screening programs, respectively. With limited resources in many areas where cancer rates are highest, the devices should be low-cost for the opportunity to reach the most communities and easy-to-operate by healthcare providers of any skill level. Advances in 3d-printing, hardware, and software technologies enable low-cost, smartphone-based medical imaging devices to be quickly developed and field tested. Integration of AFI, MSI, and polarized-white light (PWLI) imaging modalities along with machine-learning-based image classification further extends the smartphone's capabilities. Additionally, the smartphone's data transmission abilities allow the upload of images to the cloud for remote examination by specialists through web-based platforms. Presented are designs and testing results for a number of low-cost, smartphone-based imaging devices with feature sets and efficacies that rival higher-cost systems. A dual-view oral cancer screening device with remote specialist and convolutional neural network (CNN) classification achieved sensitivities, specificities, positive predictive values, and negative predictive values ranging from 81% to 94% compared to an on-site specialist's diagnosis. A second intraoral probe device improves on the previous by significantly reducing its cross-sectional area and adding a flexible section, improving patient comfort and access to significant oral cancer areas in the oropharynx and base of tongue. Lastly, two dermascopes utilizing MSI and PWLI are compared for skin cancer screening and erythema monitoring through chromophore mapping. As image databases are built and machine learning classification algorithms improve, these devices have the potential to transition from adjunctive to primary detection tools, reducing the number of biopsies and gold-standard histopathological analyses required.