Mitotic counting, or the assessment of figures indicative of cellular division, is fundamental to the pathological examination of breast cancer tissues, as it plays a pivotal role in the analysis of disease staging. Seasoned pathologists know all too well how critical precision in this step is to diagnosis, but at the same time, how labor-intensive and error-prone traditional methods can be. Given the steadily growing flow of cases in pathology labs, the pressing need for a new approach that is accurate and more efficient has never been more pronounced. In this context, the advent of artificial intelligence (AI) stands as an important ally, significantly augmenting the capabilities of pathologists in breast cancer diagnosis.
Traditional mitotic counting and Its challenges
At the very core of breast cancer diagnostics, mitotic counting demands a pathologist’s unwavering attention as they scrutinize glass slides under the microscope. The goal is to locate a hotspot, an area brimming with mitoses, and then perform a manual count of each event. Traditional mitotic counting, however, comes with a litany of challenges that can compromise its reliability. Identifying the precise hotspot is inherently subjective, often leading to discrepancies among pathologists. In fact, a recent study published in the Journal of Clinical Pathology found that pathologists often don’t agree on what they see, which can cause mistakes in determining the severity of the cancer and ultimately how it’s treated. This is because the process of counting each mitosis is not only tedious, but fraught with potential counting errors, magnified under the pressures of increasing workloads.
Another major issue with the conventional technique is the lack of standardization. Variability in microscopes, each offering different magnification and field areas, introduces an additional layer of inconsistency in the counting process. This variability can lead to significant differences in patient diagnosis and prognosis, as the mitotic index is a crucial parameter in breast cancer grading.
The rise of digital pathology and AI integration
The shift towards digital pathology has marked a significant advancement in breast cancer diagnosis. High-resolution digital imaging of slides provides pathologists with an unprecedentedly clear and expansive view of tissue samples for their analysis. The addition of digital tools, such as automated measurement, area grids, and sophisticated annotation capabilities, further enhances the accuracy and efficiency of the diagnostic process. Yet, it is the synergy of AI with these digital tools that has really initiated the most transformative shift.
AI algorithms, when layered onto digital pathology, offer a new level of precision and efficiency. These advanced applications have been designed to overcome the traditional challenges faced by pathologists. With AI, the once subjective process of identifying hotspots with human eyes and microscopes alone can be standardized, minimizing variability, and improving consistency across diagnoses. AI can systematically annotate each mitotic figure within these hotspots, supporting pathologists by ensuring no significant detail is overlooked. Moreover, these tools can automatically compute the mitotic count across entire slides and within specific hotspots, substantially easing the workload of pathologists and reducing the time taken to reach a diagnosis.
Developing effective AI tools: Key considerations
For such AI to be used in clinical practice, however, it must be underpinned by a foundation of high-quality, diverse training data. This ensures that the AI algorithms can effectively recognize and analyze the wide range of histological features encountered in various patient samples. Rigorous and ongoing testing and validation of these AI systems by practicing pathologists are essential to maintain their accuracy and clinical relevance. Furthermore, incorporating direct feedback from pathologists into the design and refinement of AI tools guarantees that these systems address the real-world demands and intricacies of the diagnostic process.
Beyond the size of datasets, scientific validity hinges on statistical significance and a demographic representation that mirrors the broader population. The medical community has long grappled with the obstacle of non-standardized data collection. This is particularly true for data on racial and ethnic disparities, which is almost absent due to inconsistent reporting levels across various health systems, insurance providers, and public health records. This is one of the key hurdles for most datasets undergoing FDA review and is why out of the more than 500 AI algorithms approved by the FDA, there is only one approved for clinical use in the field of pathology.
Embracing a new era in breast cancer diagnostics
As breast cancer diagnostics evolve, the integration of AI presents a horizon brimming with possibilities. Pathologists equipped with AI tools are already providing more precise, efficient, and swift diagnoses. This is a major step forward for a field where the speed and precision of AI can complement the nuanced judgment of experienced pathologists, creating a healthcare landscape that is not only more responsive but also more resilient. As AI continues to mature and integrate within the clinical workflow, its potential to revolutionize not just breast cancer diagnosis, but also the broader spectrum of medical diagnostics, will help ensure that every patient benefits from the innovations that promise better outcomes.
Photo: utah778, Getty Images