AI as a driving force for adoption of digital pathology workflow

Adoption of digital pathology workflow has been slow internationally, mostly due to uncertain return on investment. Compared with radiology, where a digital workflow can save costs by eliminating the need for physical films, a digital pathology workflow adds costs to traditional process. The advent of deep neural network-based artificial intelligence (AI) technology has brought hope that with its ability in assisting physicians to diagnose diseases more effectively and efficiently, adoption of digital pathology workflow can become more prevalent. Earlier research on pathology AI focused on demonstrating the capabilities of deep neural networks in recognizing lesions or pathogens. After these capabilities have been proven, it has become increasingly important to evaluate quantifiable impact AI can bring to pathology so as to facilitate adoption of an AI-powered digital pathology workflow. Here I will use our own examples to illustrate how we approach this problem. First, I will demonstrate how a metastasis detection AI model can improve diagnostic sensitivity while reducing review time. Second, I’ll show how we’ve designed a new workflow where we use a previously developed nasopharyngeal carcinoma recognition AI model to create a new review process where pathologists can achieve the same diagnostic performance while reviewing less image content. Third, I’ll show how we can use AI to gain quantitative insights into hematologic diseases that are not attainable through purely manual effort. Finally, I will conclude with directions in which we can leverage AI to implement a better diagnostic workflow for pathology.