Breast cancer grading is an important part of the diagnostic workup in breast cancer diagnostics. It guides treatment decisions by indicating how aggressive the cancer is. This article explains how the grading is conducted, its challenges, and how artificial intelligence can aid pathologists in the process.
Nottingham grading system (NGS), also known as the Elston-Ellis modification of the Scarff-Bloom-Richardson grading system, is a widely used standard for breast cancer grading. Since it was published in 1991, it has been adopted in Europe and the United States and is recommended by the International Union Against Cancer and the World Health Organization.1
In practice, if a pathologist observes invasive cancer in the specimen, they will evaluate how aggressive the cancer looks by assessing three histological features:
These three features are scored independently, with scores of 1 to 3. Based on the combined total of these scores, the pathologist then assigns a histological grade ranging from 1 to 3. The higher the grade, the more aggressive the cancer.
Several studies have demonstrated significant discrepancies among pathologists in the grading, indicating high inter-observer variability.3-7
For example, the South Sweden Breast Cancer Group conducted a study where they distributed H&E-stained slides from 93 invasive breast cancers to seven different pathology departments for blinded evaluation. The same histologic grade was obtained for all departments in 31% of the cases. The agreement was best for tubular formation and poor for nuclear pleomorphism and mitotic count.5
Lack of reproducibility and reliability of pathological examination can lead to differences in the risk assessment and treatment decisions and, thus, affect patient outcomes.8 The study of Robbins et al. (1995) suggests that experience and precise grading guidelines are valuable in improving the level of agreement between pathologists.9 However, there is no global consensus on how to perform breast cancer grading.
Artificial intelligence can solve the challenges discussed above, as it harmonizes and standardizes how breast cancer grading is performed. It can help pathologists pay attention to relevant regions, for example, when assessing the mitotic count. It also allows accurate and reproducible assessment of nuclear pleomorphism, which is currently highly subjective.
The grading performed by AI should be transparent, and pathologists should be able to understand and explain the results. Therefore, a user interface with a clear visualization of the results is pivotal in AI solutions. In general, as digitalization progresses in pathology, image analysis solutions have a vast potential to become the standard also in other diagnostic tasks because of computers' inherent high precision and repeatability compared to human readings.
Aiforia® Breast Cancer Grading is a clinical AI solution that automates breast cancer grading from H&E-stained whole-slide images (WSI), accurately identifying invasive carcinoma and ductal carcinoma in situ (DCIS). It objectively scores mitotic count, tubular formation, and nuclear pleomorphism, addressing key challenges of manual grading such as variability and time constraints, consistent with the Nottingham Grading System.
Paired with Aiforia® Clinical Suite Viewer, the user benefits from automated reporting (according to CAP guidelines), worklist sorting based on carcinoma findings, and transparent results, allowing the pathologist to review and verify the AI model output on top of the original image at high resolution.
To summarize, the clinical benefits of Aiforia® Breast Cancer Grading include:
Aiforia® Breast Cancer Grading is CE-IVD marked for diagnostic use in EU and EEA countries (IVDR) and for Research Use Only (RUO) and Performance Studies Only (PSO) in all other market areas.
Aiforia® Breast Cancer Grading is part of the Aiforia® Breast Cancer Suite, which offers a fully digital cockpit for breast cancer diagnostics, supporting the entire diagnostic workflow. Learn more about its benefits here.
References
1. Elston, C. & Ellis, I. (1991). Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: experience from a large study with long-term follow-up. https://doi.org/10.1111/j.1365-2559.1991.tb00229.x
2. van Dooijeweert, C. et al. (2022). Grading of invasive breast carcinoma: the way forward. Vinchows Arch., 480(8), 33-43. https://doi.org/10.1007%2Fs00428-021-03141-2
3. Frierson Jr, H. et al. (1995, February). Interobserver reproducibility of the Nottingham modification of the Bloom and Richardson histologic grading scheme for infiltrating ductal carcinoma. Am J Clin Pathol., 103(2), 195-198. https://doi.org/10.1093/ajcp/103.2.195
4. Sloane, J. et al. (1999, January). Consistency achieved by 23 European pathologists from 12 countries in diagnosing breast disease and reporting prognostic features of carcinomas. European Commission Working Group on Breast Screening Pathology. Virchows Arch., 434(1), 3-10. https://doi.org/10.1007/s004280050297
5. Boiesen, P. et al. (2000). Histologic grading in breast cancer--reproducibility between seven pathologic departments. South Sweden Breast Cancer Group. Acta Oncol., 39(1), 41-45. https://doi.org/10.1080/028418600430950
6. Theissig, F. et al. (1990, December). Histological grading of breast cancer. Interobserver, reproducibility and prognostic significance. Pathol Res Pract. 186(6), 732-736. https://doi.org/10.1016/s0344-0338(11)80263-3
7. Hopton, D. et al. (1989, February). Observer variation in histological grading of breast cancer. Eur J Surg Oncol., 15(1), 21-23. https://pubmed.ncbi.nlm.nih.gov/2645174/
8. Bueno-de-Mesquite, J. et al. (2010, January). The impact of inter-observer variation in pathological assessment of node-negative breast cancer on clinical risk assessment and patient selection for adjuvant systemic treatment. Annals of Oncology, 21(1), 40-47. https://doi.org/10.1093/annonc/mdp273
9. Robbins, P. et al. (1995, August). Histological grading of breast carcinomas: a study of interobserver agreement. Hum Pathol. 26(8), 873-879. https://doi.org/10.1016/0046-8177(95)90010-1