In recent years, artificial intelligence has made remarkable strides, improving various aspects of our daily lives. One notable application is in intelligent chatbots that use deep learning models. These systems have shown tremendous promise in the medical sector, enhancing healthcare quality, treatment efficiency, and cost-effectiveness. However, their role in aiding disease diagnosis, particularly chronic conditions, remains underexplored. Addressing this issue, this study employs large language models from the GPT series, in conjunction with deep learning techniques, to design and develop a diagnostic system targeted at chronic diseases. Specifically, performed transfer learning and fine-tuning on the GPT-2 model, enabling it to assist in accurately diagnosing 24 common chronic diseases. To provide a user-friendly interface and seamless interactive experience, we further developed a dialog-based interface, naming it Chat Ella. This system can make precise predictions for chronic diseases based on the symptoms described by users. Experimental results indicate that our model achieved an accuracy rate of 97.50% on the validation set, and an area under the curve (AUC) value reaching 99.91%. Moreover, conducted user satisfaction tests, which revealed that 68.7% of participants approved of Chat Ella, while 45.3% of participants found the system made daily medical consultations more convenient. It can rapidly and accurately assess a patient's condition based on the symptoms described and provide timely feedback, making it of significant value in the design of medical auxiliary products for household use.
Chronic diseases are increasingly receiving attention on a global scale. According to statistics, nearly 25% of adults suffer from one or more chronic conditions, making chronic diseases a major challenge facing global health. Not only do chronic diseases result in poor health, disability, and even death, but they also account for a substantial portion of healthcare expenditures. Therefore, the early diagnosis of chronic diseases becomes critically important. Prompt intervention is required to prevent further progression of the disease. With changes in lifestyle and the impact of the pandemic, the incidence of chronic diseases is on the rise. Consequently, the future demand for health management as well as expenditures on healthcare systems are also gradually increasing. At the same time, the demand for healthcare resources such as hospitals and doctors has also significantly increased. Patients need to spend more time waiting for medical treatment, and doctors are facing immense workloads and patient volumes. However, to reduce both time and financial costs, an increasing number of people are leaning towards contactless consultations and seeking medical advice online. In a home setting, when experiencing physical discomfort, people are more inclined to conduct online searches for medical advice. Indeed, the complexity and uncertain reliability of online information present a potential problem. Therefore, there is an urgent need for remote medical interventions in the management of chronic diseases. Monitoring provided by remote healthcare service organizations can help in the early identification of symptoms and enable rapid responses to disease exacerbation. Developing a system that can accurately assist in the diagnosis of chronic diseases becomes especially important.
Telemedicine is a popular aspect of electronic medical applications and has seen increased usage with the development of mobile healthcare technologies. This growth has established the significance and value of remote interventions in the management of health services. These technological solutions have the potential to make patients' lives easier. For example, within clinical decision support systems (CDSS), models such as naive Bayes, neural networks, and K-nearest neighbors (KNN) are utilized to diagnose COVID-19. The VP-expert diagnostic system is specifically designed for populations in underdeveloped regions, aiming to rapidly diagnose diabetes and enhance the diagnostic efficiency of physicians. The recommendation system for iron deficiency anemia, based on clinical oncology, serves as a CDSS, used for diagnosing and managing its treatment. Furthermore, the smart tracking system for acute respiratory infections utilizes mobile health technologies to infer new facts from medical data collected during examinations of pediatric patients. The use of CDSS to assist in enhancing the diagnostic efficiency of healthcare providers has become an established fact.
Currently, as the achievements of artificial intelligence in various domains have been effectively validated, deep learning has also exhibited exceptional performance in a variety of CDSS. The concept of deep learning was introduced in 2006. The primary characteristic of deep learning is unsupervised feature learning, which provides a novel solution for dealing with abstract concepts in human thought. Deep learning algorithms have been effectively applied in numerous domains, including image retrieval and web-based knowledge retrieval. For example, expert diagnostic systems for the automatic identification of asthma and chronic obstructive pulmonary disease (COPD) based on deep learning can interpret respiratory sounds recorded by a stethoscope, assisting doctors with remote access. Such systems can provide more timely and accurate diagnoses, offering a quicker and more efficient approach, especially for patients in resource-poor areas or those who cannot receive immediate medical attention. Additionally, multi-modal learning using foundational architectures like long short-term memory can also enable early predictions of exacerbations in COPD. By analyzing a range of modal data, including clinical data and radiographic information, these models capture potential disease features and trends. This helps doctors take timely intervention measures and reduce the risk of disease exacerbation. Deep learning has also shown promise in predicting hospital mortality rates. Through model training and optimization, the risk of patient mortality during hospital stays can be more accurately predicted. This is crucial for the rational allocation of hospital resources, prioritized patient care, and treatment decision-making.
As previously described, research has demonstrated the potential achievements of artificial intelligence in the realm of healthcare services. Among them, conversational agents (CAs), also known as AI chatbots, are applications capable of communicating via natural language. And with the rise of large language models, open AI's generative pre-trained transformer (GPT) has gained prominence. Trained on extensive text datasets, these models can generate natural language and be fine-tuned for a variety of language tasks. For example, tasks such as language translation, text summarization, and text completion can be undertaken. After being trained on extensive conversational text datasets, these models can understand the context and intent of a conversation. They can be utilized as virtual assistants and other conversational interfaces that generate human-like responses in dialogue scenarios. While initially mainly used for social chatting, these models are now increasingly being employed in specialized industries. Research in the field of intelligent diagnostics has demonstrated the feasibility of using large language models for radiological decision support in clinical settings, establishing them as effective tools for improving clinical workflows. The isolation measures during the COVID-19 pandemic have heightened people's awareness and understanding of online medical treatment. Doctors and patients have resorted to online consultations when face-to-face interactions are not possible. The number of remote consultations increased by 50%, further highlighting the importance of telemedicine.
GPT, as a natural language processing model, serves various roles in healthcare. For instance, it can generate radiology reports to save radiologists time or assist in diagnostic decisions by providing differential and refined information. It can even communicate with patients, offering information regarding test results and follow-up treatment recommendations. Additionally, some research has noted that digital therapy services provided by chatbots for mental health counseling have also gained recognition for their effectiveness. Therefore, the future of healthcare will largely depend on the ability to accurately perform remote diagnoses. The collection of data remotely and its analysis through artificial intelligence will contribute to improved healthcare services and health outcomes. Regrettably, there is currently a relative paucity of research focusing on chronic disease auxiliary diagnosis based on large language models. Seizing this opportunity, our research aims to develop a system for aiding in the diagnosis of chronic diseases by leveraging large language models. This aims to further enhance the quality of remote medical consultation services. This study is significant for the application of language models in healthcare and the development of systems that assist in the diagnosis of chronic diseases.
In the current technological context, this study comprehensively leverages the GPT-2 deep learning model. Through meticulous training, optimization, and packaging processes, we have successfully designed and developed an intelligent system for the auxiliary diagnosis of chronic diseases -- Chat Ella. This system is capable of engaging in dialogues with patients, deeply inquiring into related symptoms, and thereby providing preliminary diagnostic results. Such an approach enables patients to gain an initial understanding of their health conditions and consult professional doctors in a timely manner when necessary. After a series of usability tests and evaluations on a validation set, including metrics such as accuracy and AUC, the system's performance has met our anticipated standards. The chatbot-based medical auxiliary diagnosis model holds promise for broader promotion and application in the future.
The innovative aspects of this study are primarily reflected in the following areas:
