The benefits of AI Implementation in Mitotic Figure Counting

In the ever-changing landscape of pathology, the accurate assessment of mitotic activity plays a crucial role in diagnosing and managing various diseases, including cancer. Traditionally, pathologists manually count mitotic figures under a microscope in a labor-intensive and time-consuming process that is highly prone to inter-observer variability. However, with the rapid development of Artificial Intelligence (AI), we anticipate an imminent shift in these assessments, which promises significant benefits for pathologists and patients alike. In this blog post, we will explore how AI can contribute to elevating mitotic figure counting to reach new heights.

 

Understanding Mitotic Figure Counting

Before delving into the role of AI, it is essential to understand the significance of mitotic figure counting in pathology. Mitotic activity refers to the process of cell division during which cells undergo mitosis and proliferate. While this process is ever present in all living organisms, excess proliferation can lead to the growth of abnormal structures and, subsequently, cancer. The assessment of mitotic activity is one of the prognostic markers of tumor growth and carcinoma. It is typically assessed manually, under the microscope, in a subjective and labor intensive process.

 

Consistency and reproducibility of assessments

Unaided visual assessments are susceptible to various interpretations depending on the assessor’s experience, attention to detail, and expertise. Unfortunately, this inter-observer variability can lead to both over- and under-estimations of mitotic figures. Furthermore, an incorrect evaluation of proliferative cells and cancer grade can result in wrongful diagnosis and suboptimal treatment decisions.

 

Concern for misclassifications in manual pathological assessments

Despite introducing standardization methods aimed at enhancing the reproducibility of assessments, studies have revealed a relatively high misclassification rate ranging between 9% and 16% [1]. This effect tends to diminish when the sample size is above 5 mm2. The degree of misclassification can be associated with discrepancies in the microscopes utilized, contributing to even bigger problems with reproducibility across various laboratories.

Digital pathology can mitigate this inter- and intra-variability. Whole slide images (WSI) have been demonstrated to have lower variability in comparison to glass slides; however, they have been found to be more time-consuming when performed manually by pathologists [5]. AI implementation enables quick WSI analysis, increasing the reproducibility and speed of the assessments.

 

AI-enabled consistency and reproducibility

The inconsistency and inaccuracy of the mitotic counting process have been attributed to challenges in locating the most relevant mitotic hot spots, incomplete identification, and overly general classification rules [2]. Mitigation of misclassification of mitotic cells is crucial for reliable pathological assessments and research studies. With the assistance of AI, this challenge can be effectively addressed through various methods, including calculations of confidence levels, visualizations of mitotic figures, or standardizations of measurements. These practices enhance the reproducibility of lab results and ensure more precise pathological evaluations.

Figure 1 illustrates significant improvements in the accuracy of mitotic cell detection in a study comparing results with and without the use of AI technology, with different degrees of computer assistance (stages 1–3) [6]. To assess the accuracy, reviewers identified cells as mitotic or non-mitotic, and these identifications were compared with cells determined to be mitotic by ground truth (at least 4 out of 7 ground truth makers agreed). Overall, the interobserver agreement increased with AI assistance. The dispersion, or spread, of the pathologists’ assessments is notably reduced in Stage 3 (preselected region of interest and mitotic figures candidates), compared to Stage 2 (preselected region of interest only). This suggests that the additional visualizations of potential mitotic figures provided in Stage 3 resulted in a more consistent interpretation among pathologists.

Figure 1. Scatterplots comparing mitotic count values obtained from participants and algorithmic (unverified) methods against the ground truth, represented by a black line. In Stage 2, pathologists independently assessed mitotic counts within AI-selected hotspot regions. In Stage 3, potential mitotic figures in these hotspots were visualized and assessed. Both Stage 2 and Stage 3 assessments were subsequently double-checked by pathologists for accuracy. Source: Figure 4 from Bertram CA, Aubreville M, Donovan TA, et al. Computer-assisted mitotic counting using a deep learning–based algorithm improves interobserver reproducibility and accuracy. Veterinary Pathology. 2022;59(2):211-226. doi:10.1177/03009858211067478

 

Increased speed of assessments

Shortages of pathologists and their overwhelming workload

Despite over 108,000 pathologists around the world, there is a critical shortage of these medical professionals. The number of active pathologists has dropped between 2007 and 2017 by 17.5% in the USA and is dropping even further [4]. There is a need for a 45% increase in the number of residents to maintain the current turnaround. In the absence of substantial changes, patients will have to wait longer for the results of pathological evaluations, diagnoses, and treatments, which will affect healthcare as a whole. Manual mitotic counting typically takes between 5 and 10 minutes per case, which, combined with a shortage of pathologists, adds to the burden, burnout, and slower turnaround [3].

 

 Increased efficiency through AI

Studies have shown that the assessment assisted by AI can result in significant time savings compared to the assessment done solely by a pathologist [1,3,7]. A multi-center study conducted by Radboud University Medical Center showed that 75% of the participating pathologists increased their efficiency at mitosis counting/scoring when assisted by Aiosyn Mitosis Breast when compared with unassisted mitosis counting/scoring. This can allow pathologists to allocate their time and resources more efficiently and focus their expertise on interpreting results and formulating diagnostic insights.

 

Conclusion

In summary, AI has the potential to improve assessments of mitotic activity across multiple domains, revolutionize the field, and elevate patient care. By implementing AI, pathologists can achieve unparalleled efficiency, and consistency in assessing the slides, ultimately leading to more informed diagnostic and treatment decisions.

 

Aiosyn Mitosis Breast

Aiosyn recognizes the importance of improving assessments of mitotic cell activity, and therefore, we have developed a cutting-edge solution to increase efficiency and reproducibility. It facilitates the assessment of breast biopsies and resections on hematoxylin and eosin (H&E) digital slides. Currently, Aiosyn Mitosis Breast is undergoing CE certification under the EU IVDR. In the meantime, it is for research use only (RUO). Visit Aiosyn Mitosis Breast if you want to learn more about our product.

 

Sources

  1. Soliman A, Li Z, Parwani AV. Artificial intelligence’s impact on breast cancer pathology: a literature review. Diagn Pathol. 2024 Feb 22;19(1):38. doi: 10.1186/s13000-024-01453-w. PMID: 38388367; PMCID: PMC10882736.
  2. Bonert M, Tate AJ. Mitotic counts in breast cancer should be standardized with a uniform sample area. Biomed Eng Online. 2017 Feb 16;16(1):28. doi: 10.1186/s12938-016-0301-z. PMID: 28202066; PMCID: PMC5312435.
  3. Pantanowitz, L., Hartman, D., Qi, Y. et al. Accuracy and efficiency of an artificial intelligence tool when counting breast mitoses. Diagn Pathol 15, 80 (2020). https://doi.org/10.1186/s13000-020-00995-z
  4. Metter DM, Colgan TJ, Leung ST, Timmons CF, Park JY. Trends in the US and Canadian Pathologist Workforces From 2007 to 2017. JAMA Netw Open. 2019 May 3;2(5):e194337. doi: 10.1001/jamanetworkopen.2019.4337. PMID: 31150073; PMCID: PMC6547243.
  5. Al-Janabi S, van Slooten HJ, Visser M, van der Ploeg T, van Diest PJ, Jiwa M. Evaluation of mitotic activity index in breast cancer using whole slide digital images. PLoS One. 2013 Dec 30;8(12):e82576. doi: 10.1371/journal.pone.0082576. PMID: 24386102; PMCID: PMC3875418.
  6. Bertram CA, Aubreville M, Donovan TA, et al. Computer-assisted mitotic counting using a deep learning–based algorithm improves interobserver reproducibility and accuracy. Veterinary Pathology. 2022;59(2):211-226. doi:10.1177/03009858211067478
  7. van Diest, P.J., Flach, R.N., van Dooijeweert, C., Makineli, S., Breimer, G.E., Stathonikos, N., Pham, P., Nguyen, T.Q. and Veta, M. (2024), Pros and cons of artificial intelligence implementation in diagnostic pathology. Histopathology, 84: 924-934. https://doi.org/10.1111/his.15153
Share This Post

Leave a Reply