The integration of artificial intelligence (AI) in clinical pathology represents an exciting frontier in healthcare, but key challenges include data constraints, model transparency, and interoperability. These issues prevent AI and machine learning (ML) algorithms from being widely adopted in clinical settings, despite their proven effectiveness in tasks such as cell segmentation, image classification, and prognosis prediction. The benefits of AI-assisted pathology include enhanced diagnostic accuracy and efficiency, but many AI models need to be made more understandable at the single-cell level.
To address these challenges, researchers at Stanford University have developed Nuclei.io. This is a digital pathology framework that integrates active learning and real-time human feedback, designed to enhance the creation of datasets and models for various pathology applications. Two studies using Nuclei.io demonstrated its effectiveness in aiding the diagnosis of endometrial biopsies and detecting colorectal cancer metastasis in lymph nodes.
Part of the studies involved pathologists examining digitally scanned cases, using ML to improve their accuracy and reduce the need for additional tests. Nuclei.io, which was developed in Python, offers a range of features that aid collaboration between pathology and AI, providing a greater level of control and insight into ML processes.
The studies showed that by combining ML with human expertise, pathologists’ time per case could be reduced by 62.05% on average for prostate cancer diagnosis. Pathologists’ experience seemed to significantly impact the efficiency gains, with less-experienced fellows benefiting the most from the collaboration with AI. For colorectal cancer lymph node metastasis detection, ML models improved sensitivity and F1 score, particularly aiding senior pathologists.
However, the benefits varied among residents, highlighting the need for more tailor-made training and real-time adjustments. These modifications could help maintain accuracy and trust in AI technologies, and ensure they are successfully integrated into various clinical settings.
In conclusion, integrating AI and ML in the field of digital pathology holds vast potential, though the lack of data and transparent models continue to present obstacles. Nuclei.io, a Python-based software, has the potential to address these issues by accelerating ML implementation and enhancing clinician-AI collaborations. Further research is needed to refine these models and facilitate their broad application in clinical practice. This paves the way for more efficient diagnoses and improved outcomes for patients.
