Integrating Insights

Every part of the modern laboratory is integral to the development and implementation of Explainable Artificial Intelligence (XAI) in healthcare. As pathologists embrace roles that intersect more deeply with technology, their insights become essential mediators between the raw data provided by digital imaging and the nuanced, clinically relevant interpretations needed to support patient care. Championing explainability at every stage of the workflow ensures that AI tools remain accountable, understood, and trusted across all facets of medical practice.

Enhancing Outcomes

  • As laboratories become fully digital, pathologists navigate a significant transition from traditional microscopy to digital whole slide image (WSI) analyis.

  • Digitization serves as the foundation of explainability in AI.

  • The initial step in ensuring transparency and interpretability in AI-driven diagnostics begins with patient images and their data.

  • Convolutional Neural Network; CNN; Gleason Grade; Heat map; prostate cancer; prostate cancer staging;

    Driving these diagnostics are image analysis (IA) algorithms derived from deep learning (DL) networks.

    These sophisticated algorithms are crafted from a comprehensive array of WSIs and associated metadata primarily sourced from the laboratory.

  • heatmap

    A shared language is essential for bridging the gap between medical expertise and technological innovation.

    To ensure that data used to train algorithms represents the diverse patient populations they are intended to serve, development must involve collaborative inputs from both pathologists and engineers.

  • Pathologists must leverage a multitude of talents to remain at the forefront of their field.

    Computer vision, operations expertise, data science, and an innate understanding of digital technology integration are among many. Beyond traditional training, interdisciplinary collaboration is requisite for effectively managing and analyzing vast datasets, contributing to AI algorithm development, and ensuring the responsible implementation of technology in clinical practice.

  • heat map, WSI, whole slide image

    Multimodal biological data is being infused in technology-enabled clinical and research workflows throughout the globe.

    Pathologists are adopting new roles as bioinformaticians and computer visionaries of advanced medical solutions, bridging the clinical-computational divide needed to support patient care.

  • Digital pathology data plays a crucial role in training AI algorithms. The massive size of WSIs imposes significant implications for storage, financial costs, and operations. Addressing these challenges necessitates the expertise of pathologists in areas including data compression, cloud storage solutions, and efficient data retrieval systems.

  • Average weight of a humpback whale = approximate volume of traditional Hollerith “punch” cards needed to store 1gb of WSI data

  • 2x the weight of the International Space Station (ISS) = the average volume of glass slides handled yearly at major academic medical centers, large regional hospitals, and specialized reference laboratories serving multiple healthcare facilities

  • Data requirements for storing cumulative (new and archival) WSI data at a high-volume, fully digitized academic medical center (in PB) = approximately 20x the total storage capacity of typical major streaming service providers (Netflix, Amazon Prime Video, Hulu, etc.)

  • Pathologists must collaborate closely with IT specialists to optimize the infrastructure for handling these large datasets. Such ensures that the digital transition enhances rather than impedes the diagnostic process, aligning technological capacities with clinical needs. By integrating pathologist insights on workflow efficiency, data integrity, and clinical relevance, the implementation of digital pathology systems can be both cost-effective and operationally sustainable.

Pathologists are stewards of laboratory data. Diagnostic ground truth, the standard of absolute accuracy essential for training and validating AI models, is benchmarked by pathologist determination. 70% of medical decisions are based on laboratory diagnostics. The pathologist’s workflow serves as the foundation of explainability in AI. While diagnostic AI solutions for pathology today are heavily reliant on image data, the potential for incorporating a wider array of laboratory data will soon be realized.