by Joseph Chiweshe, MD, MPH, Senior Director, Medical and Scientific Affairs, Leica Biosystems
Artificial intelligence (AI) is transforming healthcare, and pathology is no exception. From tumor detection to biomarker quantification, AI-powered image analysis offers the potential for faster, more consistent insights. However, realizing this potential depends on a factor often overlooked: the quality and consistency of the inputs that feed these algorithms.
A critical question must be addressed: are the pre-analytical inputs sufficiently standardized to support reliable performance at scale?
As computational pathology continues to move from research environments into routine workflows, pre-analytic and analytic processes for slide preparation have direct implications for algorithmic performance. Variability at these stages can directly influence output quality, limiting the reliability and scalability of AI-enabled tools.
One of the foundational elements of high-quality digital pathology is the integration of advanced slide scanning technology with analytical software. Together, these components enable automated detection of common whole slide image (WSI) artifacts and support corrective actions, such as re-scanning, before images enter downstream image analysis workflows.1 When implemented effectively, these capabilities can help laboratories minimize downstream variability and maintain confidence in AI-supported outputs.
Reproducibility starts with quality control
Traditional manual quality control (QC) processes are often labor-intensive and subject to inter-operator variability, which can introduce inconsistency. These limitations may lead to delays or repeat work, affecting overall workflow efficiency.1,2
In contrast, approaches that incorporate algorithm-based QC into the scanning process allow artifacts to be identified while slides remain on the scanner, enabling earlier intervention.3 This shift—from reactive correction to proactive quality assurance—can help laboratories reduce variability at its source. Standardizing QC within the scanning workflow supports improved reproducibility and consistency in digital image generation.4
Several common histological artifacts underscore the importance of these approaches. These include air bubbles, which can obscure regions of interest and compromise interpretation; pen or ink marks, which may reduce image clarity and mask underlying tissue structures; and clipped or missing tissue, which can lead to incomplete representation of the specimen.3
Detection algorithms can identify features associated with these artifacts and flag slides for review based on predefined thresholds, enabling timely corrective action.5
Taken together, these capabilities reflect a broader shift toward integrated, data-informed workflows designed to support reproducibility and scalability. This shift is increasingly critical as AI-enabled tools move closer to clinical adoption.
Garbage in, garbage out—at scale
Variability in immunohistochemistry (IHC) is well documented. The Nordic Immunohistochemical Quality Control (NordiQC) program, which evaluates more than 600 laboratories worldwide, has reported that approximately 20–30% of submitted stains across its modules over the past two decades were graded as insufficient for diagnostic use.6
For practicing pathologists, a degree of pre-analytic variability can often be accommodated through experience and contextual interpretation. In contrast, such variability may have a more pronounced impact on algorithmic performance.
One study evaluated the robustness of an AI model using artifacts that simulate variation in tissue processing, sectioning, staining, and digitization. Across institutions and scanner types, all artifact categories were associated with a severity-dependent reduction in model accuracy.7
These findings suggest that variation tolerated in routine practice may be encoded as signal by computational models, potentially affecting performance. For laboratories implementing AI at scale, this highlights a critical reality: variability is no longer just a workflow challenge—it becomes a system-level risk to accuracy, efficiency, and confidence in results.
Minimum viable standardization
For laboratories preparing to adopt AI-enabled tools, a practical approach is to establish “minimum viable standardization.” This does not require a complete overhaul of existing workflows, but rather the implementation of a focused set of practices aimed at meaningfully reducing variability.
In practice, this approach enables laboratories to make measurable progress without disrupting established operations, aligning improvement efforts with real-world constraints.
Common areas of focus include documentation of pre-analytic variables at accession, use of assays in accordance with manufacturer instructions, implementation of routine QC using defined tissue controls and external proficiency testing, and attention to scanner calibration as digital images become integrated into the assay process.
Establishing a clear baseline of performance is a critical first step before introducing AI-based evaluation.
Foundation first, software second
Laboratories that successfully implement AI in pathology often do so by emphasizing foundational processes alongside technology adoption. This includes careful evaluation of key steps in the IHC workflow, such as tissue handling, fixation, antigen retrieval, staining protocols, and quality control measures.
In this context, technology becomes an amplifier of quality—not a substitute for it. While standardization may be less visible than downstream applications, it provides the framework that supports consistent and reliable results.
Implementation decisions remain the responsibility of the laboratory director and quality team, taking into account local workflows, validation requirements, and in accordance with manufacturer's instructions for use and applicable local requirements.
Positioning for tomorrow
Laboratories that establish standardized IHC workflows are better positioned to support the adoption of AI-based tools. Regulatory and guideline frameworks reinforce this perspective, emphasizing the importance of consistency, traceability, and validation in both traditional and AI-enabled diagnostics.
Upstream standardization represents a critical link between established quality practices and future AI readiness. Laboratories with well-defined and consistently applied processes are better positioned to scale innovation—moving from isolated implementation to sustained operational impact.
As AI continues to evolve, success will depend not only on algorithm performance, but on the strength of the systems and workflows that support it. Partnering with organizations committed to advancing workflow standardization and digital pathology can play a critical role in improving diagnostic consistency, particularly when those capabilities extend end-to-end across the continuum from biopsy to final diagnosis. Establishing this foundation today enables laboratories to translate scientific advancement into reliable, real-world outcomes tomorrow.
About the author
Joseph Chiweshe, MD, MPH is Senior Director, Medical and Scientific Affairs at Leica Biosystems. With more than 15 years of experience across the healthcare and medical device space, he is focused on active partnership with external stakeholders and communication of evidence for the medical and scientific community. Prior to Leica Biosystems and Danaher, he worked in various healthcare-related settings spanning venture capital, clinical trials, and health systems and quality administration. A graduate of the University of Nebraska, Dr. Chiweshe received his medical degree from the University of Nebraska Medical Center and Masters in Public Health from the University of Kentucky. He completed his internship in General Surgery and residency in Preventive Medicine.