An End-to-End Artificial Intelligence Eco-System to Streamline AI Applications in Wound Care

Abstract Body: Introduction: From speech recognition to detecting traffic signs, AI-driven approaches have enabled smarter technologies in different aspects of our lives. However, such advances are seldom observed in the medical field. While there is no shortage of motivation for health providers to improve on the status quo of diagnosis and treatment, access to a functional machine learning model remains difficult. Even if a Data Scientist has developed an accurate AI model, making it “consumable” by a healthcare professional requires customized interface development. We propose an end-to-end AI eco-system to give caregivers access to the powerful AI world.   Method: We developed a web platform where users go through the 4 steps of an AI project—data gathering, data labeling, model generation and model deployment. The front-end which utilizes HTML, CSS, Javascript and Vue.js has the look and feel of any web platform to provide a familiar environment. We developed an AI engine based on Python and Django which takes historical data and a prediction target as input and returns an AI model as an output, which can then be deployed right away to predict medical tasks. We compared a wound etiology predicter with the automated AI model to evaluate performance.    Result: Compared to the wound etiology classifier, the AI model generation took 2 hours to train and deploy, instead of three weeks where a Data Scientist and a web developer work in tandem. However, due to generalization requirements, the accuracy of the automatically generated model was 90% compared to the 94% developed by a Data Scientist.    Conclusion: We developed an end-to-end AI eco-system to democratize AI utilization in wound care. The positive feedback from early adaptors encouraged us to develop further functionalities. Besides opening the AI world to the health care field the development of AI models is a matter of hours now.