The same reasons that motivated radiology and cardiology to adopt DICOM standards and PACS are motivating digital pathology adoption:

• Improved clinical productivity by speeding and automating workflow
• Consolidating specialty-specific solutions to a single enterprise solution improves the economies of scale, and can lower purchase and maintenance costs
• Gaining the freedom to choose the most competitive, vendor-neutral products (scanners, storage, viewers, reporting) based on open standards. Standards-based products, enable future-proof solutions, and avoid disruptions when choosing replacement products, or to incrementally scale up and expand the capacity of the pathology practice
• Clinical collaboration and sharing data across all the institutions and clinical specialties within the enterprise

Open Standards for Pathology

Pathology images were added to the DICOM Standard (Supplement 145) more than a decade ago. WSI has unique encoding challenges: high resolution of huge size and with spatial resolution than can vary within the image, a Z plane dimension, and color or multi-spectral pixels. Over the last few years at least twelve vendors offering scanners, image management archives, viewers, and analyzers products have demonstrated interoperability by participating in digital pathology Connectathons organized by the Integrating the Healthcare Enterprise (IHE) held at conferences sponsored by Digital Pathology Association and the European Congress of Digital Pathology.

The Influence of Machine Intelligence

In recent years, Artificial Intelligence (AI) based on Machine Learning (ML) is transforming medical imaging. Machine intelligent tools can assist the medical imaging specialist perform complex image analysis tasks, classify/identify features, quantize findings, annotate the images, and report findings. These capabilities add yet more productivity gains with the potential to significantly improve quality of practice. Today there are well over 100 FDA cleared AI algorithms in radiological imaging.

AI-ML has increased the urgency toward digital solutions based on industry standards. In response, standards development organizations have been working vigorously to extend the standards to accommodate AI actors and viewers, particularly in the area of communicating AI results and annotations in clinical workflow. This article How will AI results affect medical imaging summarizes many of these standardization efforts.

In computational pathology, as pathologists refer to digital and AI analytic tools, machine algorithms annotate whole slide microscopy images (WSI) to identify features of diagnostic importance to the pathologist. The event prompting this blog, is the recent release for public comment of a pathology annotation standard that enables computational pathology applications to participate in pathology workflow, so that machine intelligent applications and viewer products can use a single standard to display images to the pathologist with their image annotations on any compliant viewer. An overview presentation by the DICOM Pathology Working Group is available here.

Annotations are points, geometric shapes, or measurements. Annotations can also carry semantics that identify cell types, cell structures, and properties such as abnormalities using coded concepts from SNOMED International, which was conceived and developed by the College of American Pathologists. For relevant examples to WSI, see SNOMED cell concepts. With coded concepts, annotations and the images they annotate, become valuable beyond the clinical setting, extending into research such as pathomics and to industry, to gather truth and images for training and retraining deployed AI models.

Moreover, annotated images are important to validate performance in new clinical settings and to monitor real world performance (RWP) during clinical use to ensure stability of the AI model and detect performance degradation caused by drift in images from the scanner. Products that depend on proprietary data are disadvantaged relative to those adhering to the industry standard. Truth data and continued performance monitoring are critical to the acceptance and success of AI in diagnostic imaging. The selection and management of training data is prominent in the Good Machine Learning Practice (GMLP) regulatory framework for Software as a Medical Device (SaMD) now under development by the FDA. See the FDA SaMD regulatory framework discussion paper and the FDA’s recently published action plan for AI and ML in SaMD

How to Prepare

• Product Managers and development engineers can improve product value (interoperability and future-proof customer data) by planning to align to standards as they develop. They can also influence the standard to meet product requirements.
• Pathologists and clinical IT managers should consider standards as they plan to purchase standards-compliant, interoperable products they can integrate into their enterprise imaging solutions.
• Get Involved. Standards depend on experts from industry, product development, and clinical experts. Participate in DICOM WG26 and the IHE Pathology and Laboratory Medicine (PaLM)