Machine learning (ML), deep learning, and artificial intelligence (AI) have a rich history in the field of medicine with its foundations being traced back to the 1950s and 1960s when researchers like Drs. Hunt, Nillson, and Rosenblatt pioneered its beginnings.1 These early developments laid the groundwork for the integration of AI and ML techniques in the field of medicine and over the years, significant advancements and breakthroughs have led to the potential of ML in healthcare being realized and brought to fruition.2 ML has been applied to various medical problems, such as computer-aided diagnosis, disease detection, and personalized medicine.2,3 Furthermore, the use of AI and ML in medicine has the potential to automate tasks, improve patient outcomes, and advance global healthcare.2,4 However, there are also challenges and considerations that remain to be addressed, such as fairness, privacy, interpretability, and ethical implications.5
Deep learning, a subset of ML, experienced a significant breakthrough in the late 2000s and early 2010s. In medicine, deep learning techniques have revolutionized areas such as medical imaging analysis, natural language processing, genomics, and drug discovery with models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) having achieved human-level or even superior performance in tasks like radiology image interpretation, pathology analysis, and clinical language understanding.6–15 These models have demonstrated exceptional capabilities in various medical imaging tasks, including the detection of fractures, classification of breast cancer histopathological images, prediction of cardiovascular events, and analysis of radiology reports.6–15 The use of deep learning techniques, particularly CNNs and RNNs, has revolutionized medical image analysis and natural language processing in the healthcare domain.6,7,11,13 These models have shown promising results and have the potential to enhance diagnostic accuracy, improve patient outcomes, and optimize healthcare delivery.6–15
During the course of the COVID-19 pandemic caused by the SARS-CoV-2 virus, extensive endeavors have focused on harnessing large datasets and deep learning techniques to facilitate the identification of individuals who have contracted the virus.16–18 These efforts have been driven by the need for quick and accurate diagnosis, especially considering the limitations of traditional diagnostic techniques in handling the exponential propagation of infection.16 ML and DL approaches have been widely employed to assist the healthcare sector in providing precise and efficient COVID-19 diagnosis using pre-existing medical imaging modalities.16–18 Unfortunately, pre-existing AI-driven imaging modalities (such as CT scan and X-ray) for the AI-guided diagnosis of COVID-19 require affected individuals to enter a medical setting to be tested, which puts both them and healthcare workers at risk of exposure.19 To date, there are no remote imaging options available outside of a medical setting to screen people in their homes.19 While cellular retinal imaging has shown promise, it typically requires the expertise of trained technologists and is challenging to perform remotely and accurately.20 Thus, the objective of this study was to investigate the potential of external eye images captured using a smartphone camera in COVID-19 diagnosis, as such images can potentially be utilized by deep learning models to exploit distinctive changes in conjunctival tissue, thereby enabling the remote diagnosis of COVID-19.
Despite the resolution of the COVID-19 pandemic as well as the progress made in vaccination efforts, the global impact of the virus continues to present significant challenges in waste management, loss of life, and global burden of disease.21–29 The pandemic has led to changes in waste generation and composition, including increased medical waste, personal protective equipment (PPE) waste, and plastic waste.23–26 These changes have put pressure on waste management systems, leading to issues such as waste stockpiling and the need for emergency treatment and disposal.22,25 These lingering consequences of COVID-19, including the continuation of preventable deaths, highlight the urgent need for alternative diagnostic approaches. Exploring novel, cost-effective, and sustainable diagnostic methods that can mitigate financial burden, reduce waste, and ensure timely and accurate identification of infectious diseases is crucial for safeguarding public health and preventing unnecessary loss of life in a post-pandemic era.
The integration of deep learning principles and the utilization of cellular camera imaging in this study represent significant advancements in the quest for improved COVID-19 diagnostics. While the research provides valuable insights, additional validation and refinement are necessary to establish the efficacy and reliability of this approach. By presenting these initial findings, this study contributes to the ongoing efforts to explore non-invasive imaging modalities and harness the potential of DL algorithms for enhancing early detection and diagnosis of infectious diseases.