Eye of the AI storm: Exploring the impact of AI tools in ophthalmology
2023; Medknow; Volume: 71; Issue: 6 Linguagem: Inglês
10.4103/ijo.ijo_1478_23
ISSN1998-3689
Autores Tópico(s)COVID-19 diagnosis using AI
Resumo"Artificial Intelligence will have a more profound impact on humanity than fire, electricity and internet." - Sundar Pichai Without us being consciously aware of it, artificial intelligence (AI) has deeply permeated every aspect of our lives – from online advertisements and movie recommendations that pop up based on our user habits, text autocorrect function on the mobile telephone, and face-recognition-based airport entry, to new generation fundus imaging systems - are all AI-based. We had invisible AI in our lives for decades, but several recently introduced tools have provided us with ready access to AI, and the ability to interact with and get the response from it in the language that we are accustomed to. Recently introduced ChatGPT by OpenAI[1] has caused great drama and excitement in the AI space. Other open access free or paid interactive AI tools such as Microsoft Bing AI, Google Bard AI, Perplexity AI, Jasper Chat, Chatsonic, Pi, GitHub Copilot X, Amazon Codewhisperer[2] have added to the variety. The AI floodgates are now open, the genie is out of the bottle, and life will never be the same. AI Tools – Illuminating Ophthalmology's Path to Precision "Intelligence is the ability to adapt to change." - Stephen Hawking The marriage of medicine with AI tools has ushered in a new era of endless possibilities. The transformative impact of these technologies on patient care and research is exemplified by their ability to provide information, facilitate diagnosis, and support healthcare professionals. Ophthalmology, being an image-based specialty, AI has a predominant role to play. As we delve into this convergence, we uncover a landscape where the AI tools could seamlessly coalesce with the existing systems, augment human expertise and propel ophthalmology into uncharted territories. There are several useful AI tools that have been specifically developed for applications in ophthalmology: Fundus Image Analysis: AI algorithms can analyse fundus images to detect and diagnose diabetic retinopathy, age-related macular degeneration (AMD), glaucoma, and retinopathy of prematurity. These algorithms can help automate the screening process, prioritize high-risk cases, and assist ophthalmologists in providing timely and accurate diagnoses. Optical Coherence Tomography (OCT): AI algorithms can analyse OCT images to identify and quantify structural abnormalities, track disease progression, and assist in diagnosing conditions such as macular oedema, AMD, and glaucoma. Automated Refraction: AI-powered devices can perform automated refraction, use machine learning algorithms to analyse wavefront measurements, and provide accurate and objective results. Virtual Assistants and Chatbots: AI-powered virtual assistants and chatbots can provide information, answer questions, and assist patients and healthcare professionals in various tasks. These tools can help with symptom assessment, provide educational resources, offer guidance on medication usage, and schedule appointments. Surgical Planning and Guidance: AI algorithms can analyse preoperative data and images to assist surgeons in planning ophthalmic surgeries. They can provide insights into optimal incision placement, lens selection, and other important considerations. During surgery, AI tools can provide real-time guidance, aiding surgeons in performing procedures with greater precision and accuracy. Disease Progression Prediction: AI models trained on longitudinal patient data can help predict the progression of eye diseases, such as glaucoma or AMD. By analysing various risk factors, patient characteristics, and clinical data, these models can provide personalized predictions, allowing for more proactive and targeted interventions. Integration of AI with diagnostic and image acquiring devices has been a value-addition and has resulted in improved diagnostic accuracy. Some of the examples are: Optovue iWellness: Optovue iWellness is an AI tool used for screening and diagnosing retinal diseases, including diabetic retinopathy and AMD. It combines optical coherence tomography (OCT) imaging with AI algorithms to provide early detection and assessment of ocular conditions. IDx-DR: IDx-DR is an FDA-approved AI tool for the autonomous detection of diabetic retinopathy. It utilizes a deep learning algorithm to analyse retinal images and provide diagnostic assessments without the need for a human interpreter. CIRRUS HD-OCT with AngioPlex: CIRRUS HD-OCT with AngioPlex is an advanced AI tool that combines OCT imaging with angiography to provide detailed analysis of retinal structures and blood flow, Eyenuk EyeArt: Eyenuk EyeArt is an AI tool that assists in the screening and detection of diabetic retinopathy. It analyses retinal images and provides automated assessments, helping to identify patients who require further evaluation by an ophthalmologist. Topcon Harmony with OphtAI: Topcon Harmony is an AI-powered software platform that integrates multiple imaging modalities, including OCT and fundus photography. It offers comprehensive image analysis and management capabilities, facilitating efficient diagnosis and follow-up care. Most of the AI-integrated diagnostics are expensive and may be readily available only at larger eye care facilities and institutions. Open AI tools, however, can be accessed by most of us free of charge or with a nominal subscription fee and can be used for the development of customized applications relevant to us. The advent of open-access AI tools has paved the way for collaboration, democratizing access to cutting-edge technology and knowledge. Here are some of the examples of open access AI tools relevant to ophthalmology: RetinaNet: An open-source deep learning framework for object detection in images, used for detecting and localizing lesions in retinal images, such as diabetic retinopathy, AMD, and glaucoma. DeepDR: A deep learning-based system developed for automated detection and classification of diabetic retinopathy in retinal fundus images. It is an open-source tool that uses convolutional neural networks (CNN) for the analysis of retinal images and provides an automated grading of diabetic retinopathy severity. Ilastik: An open-source software suite that enables interactive and trainable image segmentation. It can be used for segmenting different structures within retinal images, such as blood vessels or lesions. It provides a user-friendly interface and supports various machine learning algorithms. VGG-16: A CNN architecture applied in ophthalmology for the classification and diagnosis of retinal diseases, including diabetic retinopathy, AMD, and glaucoma. Optic Disc Segmentation Tool: An open-source software package that helps in segmenting the optic disc from retinal images. Accurate segmentation of the optic disc is crucial for glaucoma diagnosis and monitoring. OpenDR: An open-source software library for processing and analysing retinal images. It provides a range of functionalities, including image pre-processing, feature extraction, and classification. OpenDR can be used for various tasks, such as the detection of lesions, optic disc analysis, and image registration. TensorFlow: An open-source machine learning framework that provides a wide range of tools and resources for developing AI models in ophthalmology. TensorFlow.js: A JavaScript library that allows training and deploying machine learning models directly in web browsers, enabling the development of interactive web applications for ophthalmic image analysis and visualization. PyTorch: An open-source deep learning framework that offers a flexible platform for developing AI models in ophthalmology. OpenCV: A free and open-source computer vision library that includes various algorithms and functions for image and video processing in ophthalmology. Keras: An open-source neural network library written in Python, widely used in ophthalmology for tasks such as image classification, segmentation, and generative modelling. DeepLabCut: An open-source toolbox for marker-less pose estimation of animals, which can be utilized in ophthalmic research involving animal models. MedPy: An open-source library specifically designed for medical image processing tasks, including segmentation, registration, and feature extraction in ophthalmic images. MXNet: An open-source deep learning framework that provides a flexible and efficient platform for developing AI models in ophthalmology. Caffe: An open-source deep learning framework known for its speed and efficiency, commonly used in ophthalmology for image classification and object detection tasks. SciKit-Learn: A machine learning library in Python that provides a range of tools for data pre-processing, model selection, and evaluation in ophthalmic research. DeepGauge: An open-source framework for benchmarking the performance and interpretability of deep learning models, useful for evaluating AI models used in ophthalmology. CellProfiler: An open-source software for high-throughput image analysis of biological samples, including ophthalmic images, allowing for quantitative measurements and feature extraction. MXNet GluonCV: A toolkit built on top of MXNet that provides pre-trained models and tools for computer vision tasks in ophthalmology, such as image classification and object detection. OpenAI's GPT: A language model developed by OpenAI, which can be used in various applications within ophthalmology, including information retrieval, patient education, and answering specific ophthalmic queries. AI Tools to Create Videos and Presentations "There are no great limits to growth because there are no limits of human intelligence, imagination, and wonder." - Ronald Reagan AI-assistance in creating presentations and videos could be invaluable to a busy academic ophthalmologist. There are several AI-based tools that can help: Lumen5: Lumen5 is an AI-powered video creation platform that converts text-based content into engaging videos. It uses natural language processing to analyse the text, automatically generate video scenes, select appropriate visuals, and add background music. It is a useful tool for transforming blog posts, articles, or presentations into video format. Powtoon: Powtoon is an online platform that allows users to create animated videos and presentations using a drag-and-drop interface. It provides a variety of pre-designed templates, characters, and props, along with AI-powered automation features that can assist in streamlining the video creation process. Visme: Visme is a versatile visual content creation tool that includes features for creating presentations, infographics, videos, and more. It offers a wide range of templates, graphics, and animations, and incorporates AI features to help with design suggestions and content creation. Prezi: Prezi is a presentation software that utilizes AI-driven features to create dynamic and visually engaging presentations. It offers features like Smart Structures, which can automatically arrange and align the content, and Prezi Video, which enables overlaying of video onto presentation slides. It allows for a more interactive and non-linear approach to presentations. SlideBot: SlideBot is an AI-powered presentation tool that uses machine learning to analyse the content of your slides and provide real-time suggestions for improvements. It can help optimize slide layouts, text placement, and font choices to create visually appealing and effective presentations. Microsoft Office AI: Copilot is a modern AI assistant that comes with Microsoft Office that will help create stunning documents on Word, presentations on PowerPoint and robust worksheets on Excel. DALL-E: DALL-E and its iterations are deep learning models developed by OpenAI to generate lifelike digital images from natural language descriptions, called "prompts". Figs. 1-5 are some of the examples of images created by Dall-E using simple prompts. Figure 1: An oil painting of an ophthalmologist at work (what Michelangelo would have created, imagined and generated by DALL-E)Figure 2: An oil painting of an ophthalmologist (what MF Hussain would have created, imagined and generated by DALL-E)Figure 3: Raja Ravi Varma may have depicted an ophthalmologist like this (generated by DALL-E)Figure 4: Painting of an eye similar to what van Gogh may have painted (generated by DALL-E)Figure 5: Ophthalmology in the year 2200 imagined and (generated by DALL-E)ChatGPT: Empowering Ophthalmology's Conversations with Artificial Intelligence "A computer would deserve to be called intelligent if it could deceive a human into believing that it was human." - Alan Turing ChatGPT (chat = chatbot functionality and GPT = Generative Pre-trained Transformer), launched on November 30, 2022, is an AI tool, a sort of a chatbot on steroids, that is endowed with the ability to understand and process natural language and provide human-like responses to text input.[1] It has over 175 billion parameters and currently gets over 10 million user queries every day. It was trained on over 570 GB of data, including web pages, books, and other sources of knowledge. It has generated tremendous interest in AI, garnered over 100 million users in less than 6 months surpassing Google+ (which took over a year to reach that milestone), and has spawned hundreds of easy-to-use applications.[3] ChatGPT is emerging as a possible game-changer. Its potential as an information resource is immense. Beyond information dissemination, ChatGPT's role in symptom assessment and triage is crucial. Here are some specific ways ChatGPT can be beneficial: Information and Education: ChatGPT can provide general information about eye anatomy, common eye conditions, diagnostic procedures, and treatment options. It can help patients and healthcare professionals understand various eye-related topics and provide educational resources. Symptom Assessment: ChatGPT can assist in evaluating symptoms and providing preliminary guidance. Users can describe their eye-related concerns, and ChatGPT can ask relevant questions to gather information and offer possible explanations or next steps. Patient Triage: In certain scenarios, ChatGPT can assist in triaging patients by asking questions about their symptoms and guiding them toward appropriate care. It can help identify urgent cases that require immediate attention and provide guidance on whether the patient should seek emergency care or schedule an appointment with an ophthalmologist. Medication Information: ChatGPT can provide information on various eye medications, including their uses, dosages, and potential side effects. Lifestyle Recommendations: ChatGPT can offer general advice on maintaining good eye health, including recommendations for proper nutrition, eye hygiene, eye protection, and strategies for reducing eye strain from digital devices. Preoperative and Postoperative Support: ChatGPT can provide information and answer common questions related to ophthalmic surgeries, such as cataract surgery, LASIK, or retinal procedures. It can offer guidance on preoperative preparations, postoperative care, and recovery timelines. Apart from its evident potential utilities in patient care, ChatGPT can be utilized in research in ophthalmology in several valuable ways: Data Analysis and Mining: ChatGPT can assist in analysing large volumes of textual data related to ophthalmic research. Researchers can use ChatGPT to extract and summarize key information from scientific articles, clinical trial reports, electronic health records, and other relevant sources. This can help in identifying patterns, trends, and associations within the data, enabling researchers to gain insights and support their investigations. Literature Review and Knowledge Synthesis: ChatGPT can aid researchers in conducting literature reviews by providing summaries of existing research articles, identifying knowledge gaps, and suggesting potential areas for further exploration. Researchers can interact with ChatGPT to ask specific questions related to their research interests, and it can help in synthesizing relevant information from the existing literature. Hypothesis Generation and Exploration: ChatGPT can be a useful tool for brainstorming and generating new research ideas. Researchers can engage with ChatGPT to discuss their hypotheses, explore alternative explanations, and refine their research questions. ChatGPT's ability to generate diverse responses can assist in stimulating creative thinking and identifying novel perspectives in ophthalmic research. Collaboration and Peer Review: ChatGPT can facilitate collaboration among researchers by providing a virtual platform for exchanging ideas, discussing methodologies, and seeking feedback. Researchers can use ChatGPT to engage in scientific discussions, present their findings, and receive input from their peers. This can help improve the quality of research projects and foster a collaborative environment within the field of ophthalmology. Patient and Public Engagement: ChatGPT can be leveraged to engage with patients and the public in ophthalmic research. It can help researchers gather information on patient experiences, preferences, and perspectives through interactive conversations. ChatGPT can also provide educational resources and answer questions about ongoing research studies, informed consent, and the potential impact of research findings on patient care. Ethical Considerations and Guidance: ChatGPT can support researchers in navigating ethical considerations related to ophthalmic research. It can provide information on research ethics, data privacy, and informed consent. Researchers can interact with ChatGPT to discuss ethical dilemmas, explore different viewpoints, and seek guidance on conducting research in a responsible and ethical manner. How Smart is Smart? Testing ChatGPT "We should take care not to make the intellect our goal; it has, of course, powerful muscles, but no personality." - Albert Einstein The efficacy and reliability of ChatGPT in performing several tasks related to ophthalmology are being anxiously explored. Its ability to search references, write an academic article, format operating notes and discharge summaries, provide patient information, answer curriculum-based examinations etc., have been tested.[4–14] Valentín-Bravo et al.[6] used ChatGPT to generate an abstract and a structured article, title and references on the topic silicone oil in vitreoretinal surgery. The authors concluded that despite the knowledge demonstrated ChatGPT, the scientific accuracy and reliability on specific topics is currently insufficient for the automatic generation of scientifically rigorous articles. Mihalache et al.[11] have recently tested ChatGPT's ability to answer practice questions for board certification in ophthalmology and found that it could answer about 46-58% of questions correctly - encouraging, but not good enough yet. While its ability to write robust academic articles is still rudimentary at best, the ethical aspects of using ChatGPT in academics are already being fervently questioned, and rightly so.[14] Guidelines and ethical boundaries need to be established to define the exact role of AI tools in abstract and manuscript preparation. At this point in time, it is considered inappropriate to use ChatGPT or any such AI tool to generate scientific abstracts and complete academic articles. There are tools to detect AI-generated text and extensive use of such text may a justification for manuscript rejection. A major disadvantage of ChatGPT is that it has been trained only until September 2021. However, updates are being released and it will hopefully become up-to-date. I entered into a prolonged conversation with ChatGPT on a diverse range of topics related to ophthalmology (see box). Robust information that it provides, dexterity in providing straight answers to convoluted questions, advanced language skills, and the ability to compose creative poems on mere prompts are mind-boggling. Conclusion "These little grey cells. It is up to them."- Agatha Christie AI is all-pervasive and has already deeply permeated several aspects of medical care. AI tools are expected to play an increasingly significant role in revolutionizing ophthalmic care. Its potential beneficial role in improving accuracy in diagnostics, optimizing workflows, building effective patient education and counselling methodologies, establishing optimal patient triage and management pathways, improving safety and precision in surgery, and in fact, revolutionising the entire patient care ecosystem is undeniable. AI tools and intelligent automation can augment the capabilities of ophthalmologists by providing a third hand, a watchful eye, an attentive ear, a tireless brain with an unlimited storage capacity, and in all, an uncompromising ability to minimise errors and optimise patient care. Recently introduced natural language interactive AI tools such as ChatGPT have provided us with a peep window to the AI world. It is for us to explore the endless possibilities to the fullest and build synergy between the mind and the machine for the good of all. "Every great advance in science has issued from a new audacity of imagination." - John Dewey
Referência(s)