Streamlining the generation of AI tools on a cloud medical imaging platform

Artificial intelligence algorithms have made undeniable inroads into the medical imaging field. In Digital Pathology, AI will play a key role in enhancing productivity by decreasing image screening time and improving the quality of diagnoses. AI tools can learn complex image patterns, which enable them to segment, and subsequently classify them. The adaptability feature proves beneficial across various medical imaging modalities. Despite that, the development of these algorithms requires the availability of substantial labeled data, which is often inaccessible while demanding efforts from trained personnel for the annotation process and also data scientists to develop the algorithms. These requirements often put such algorithms beyond the reach of doctors and medical specialists. Therefore, to simplify the process of creating image analysis tools, this work introduces an architecture that directly integrates AutoML technologies into a web-based PACS system to allow users without prior experience to build their own classifier or model. The system underwent validation within a digital pathology context, typically characterized by extensive data volumes, specifically for the annotation of mitotic cells. The achieved results were benchmarked against current state-of-the-art methods, with an F1-score of 0.81 versus an average of 0.82 between the research analyzed.