Ibrahim, Asmaa
(2024)
Advancing proliferation assessment in invasive breast cancer: integrating digital pathology, artificial intelligence, and molecular insights for improved diagnostic and prognostic strategies.
PhD thesis, University of Nottingham.
Abstract
Background:
Mitotic score is one of the three components of the Nottingham grading system of breast cancer (BC), and it represents the tumour’s rate of proliferation and aggressiveness. However, the optimal method of mitoses assessment in BC remains to be defined. Digital pathology (DP), combined with artificial intelligence (AI) algorithms, offers solutions for the challenges associated with the manual counting of mitotic figures. AI algorithms can improve the detection and quantification of mitotic figures. However, the lack of consistent methodology and definitions hinders their integration into clinical practice.
In addition to sustained proliferation, the evasion of apoptosis is one of the prerequisites for carcinogenesis and tumour progression. High levels of apoptosis may appear to be a good prognostic factor; however, apoptosis is frequently increased in high-grade tumours, particularly when it is accompanied by high levels of proliferation, and the evidence for apoptosis as an independent prognostic marker remains inconclusive. This suggests that the regulators that control proliferation and apoptosis in normal tissues persist, at least in part, during malignancy. This can in part justify why these two parameters, mitosis, and apoptosis, should be analysed together; the targeting of their regulatory mechanisms could offer the opportunity to discover new therapies and to identify potential drug targets.
Genes associated with cell proliferation have the most powerful prognostic potential. Mitosis, as the active phase of cell division, offers an objective and accurate measurement, surpassing other proliferation indices. This critical process is regulated by specific gene drivers in BC. Therefore, it is crucial to explore candidate genes specifically associated with effective mitotic cell division, as determined by mitotic scores, to obtain a more accurate representation of dividing cells in BC. Such exploration will lead to a deeper comprehension of BC pathogenesis and the development of novel therapeutic targets and personalised treatment strategies.
This PhD study aimed to improve the assessment methods of mitotic counts using DP and whole slide image (WSI) technology, evaluate the role of AI in improving the performance of mitotic count estimation and improve the understanding of molecular markers associated with mitotic activity in BC.
Methods and results:
To determine the optimal method used to count mitotic figures in WSI, digital images of 595 haematoxylin and eosin (H&E) stained sections of BC were evaluated. Several morphological criteria were investigated and applied to define mitotic hotspots, while reproducibility, representativeness, time, and association with outcome were the criteria used to evaluate the best area size for mitosis counting. Three approaches for scoring mitoses on WSIs were evaluated: single annotated rectangles, multiple annotated rectangles, and multiple digital screens of screen fields (HPSF). The findings indicated that the relative increase in tumour cell density was the most significant parameter for identifying mitotic hotspots. Counting mitoses in a 3mm2 area was the most representative, regarding saturation and concordance levels. Counting in an area less than 2mm2 resulted in a significant reduction in mitotic count (p=0.02), whereas counting in an area ≥4mm2 was time-consuming and did not produce a significant rise in the overall mitotic count (p=0.08). The use of multiple HPSFs, following calibration, provided the most reliable, time-saving, and practical method for mitoses counting on WSI.
To assess the value of apoptosis when considered in the context of mitotic activity, apoptotic and mitotic figures were counted in WSI generated from H&E-stained sections of 1,545 BC cases derived from two well-defined BC cohorts. Counts were carried out visually within defined areas. The findings indicated a significant correlation between mitotic and apoptotic scores. High apoptotic counts were associated with features of aggressive behaviour, including a high grade, a high pleomorphism score and hormonal receptor negativity. Although the mitotic index (MI) and apoptotic index were independent prognostic indicators, the prognostic value was synergistically higher when combined. BC patients with a high combined apoptotic index and MI had the shortest survival. The replacement of the mitosis score with the mitosis-apoptosis index in the Nottingham grading system revealed that the grade with the new score had a higher significant association with breast cancer-specific survival (BCSS) with a higher hazard ratio.
To assess whether mitotic counts can be improved using immunohistochemistry, two full-face sections from 97 cases were cut, one stained with H&E only, while the other was stained with Phosphohistone H3 and counterstained with H&E (PHH3-H&E). The process of counting mitoses using PHH3-H&E was compared to traditional mitoses scoring using H&E in terms of reproducibility, scoring time and the ability to detect mitosis hotspots. The agreement between the manual and image analysis-assisted scoring of mitotic figures using H&E and PHH3-H&E-stained cells was assessed, and the diagnostic performance of PHH3 in detecting mitotic figures in terms of sensitivity and specificity was measured. Finally, PHH3 was replaced with mitosis score in a multivariate analysis to assess its significance. Significantly higher mitotic figures were detected using the PHH3-H&E when compared with H&E alone (p<0.001). The concordance between pathologists in identifying mitotic figures was highest when using the dual PHH3-H&E technique. In addition, it also highlighted mitotic figures at low power, allowing better agreement on choosing a hotspot area in comparison with standard H&E. A better agreement between image analysis-assisted software and the human eye was observed for PHH3-stained mitotic figures. When the mitosis score was replaced with PHH3 in a Cox regression model with other grade components, PHH3 was an independent predictor of survival (p=0.002) and even showed a more significant association with BCSS than mitosis (p=0.005) and Ki67 (p=0.27).
To identify differentially expressed genes associated with mitotic activity, WSIs generated from H&E-stained sections of The Cancer Genome Atlas (TCGA) BC database (n=1053) were utilised, alongside their transcriptomic and clinical data, and the mitotic figures were counted to stratify them into high and low mitotic score sets. Genes enriched in the cell cycle pathway were utilised to predict the protein–protein interaction network. Ten hub genes (ORC6, SKP2, SMC1B, CDKN2A, CDC25B, E2F1, E2F2, ORC1, PTTG1, CDC25A) related to mitotic scores were identified using CytoHubba, a plug-in in Cytoscape. In a multivariate Cox regression model, ORC6 and SKP2 were predictors of survival independent of existing methods of proliferation assessment, including mitotic score and Ki67. The prognostic ability of these genes was validated using the Molecular Taxonomy of Breast Cancer International Consortium (n=1980) and a combined multicentric cohort (n=7303). Finally, the protein expression of these two genes was validated on a large cohort of BC cases, and they were significantly associated with poor prognosis and patient outcome. A positive correlation between ORC6 and SKP2 mRNA and protein expression was observed.
To further investigate the expression of ORC1 in BC, ORC1 mRNA and protein expression was evaluated and correlated with clinicopathological parameters and patient outcomes. High ORC1 mRNA and protein expression were found to be associated with poor prognostic features, such as larger tumour size, high tumour grade and nodal stage and poor outcomes. Additionally, the combining of ORC1 and ORC6 mRNA expressions independently predicted shorter BCSS.
In evaluating the key proteins involved in mitosis that could play important roles in BC progression and prognosis, INCENP, ASPM, ESPL1, SPC25 and DSN1 were identified. These proteins were found to have potential as prognostic biomarkers in BC and may have clinical relevance in predicting survival rates and treatment response. The targeting of these proteins could also be a potential therapeutic strategy for managing BC and other tumour types characterised by their overexpression.
Finally, to facilitate the implementation of AI BC mitotic assessment, WSIs from a large cohort of BC with extended follow-up (Nottingham cohort), comprising a discovery (n=1715) and a validation (n=859) set, were utilised. The TCGA BC cohort (n=757) was used as an external test set. Employing automated mitosis detection, three methods were tested: the mitotic count per tumour area (MCT), the MI and the mitotic activity index (MAI); the best approach identified was also compared with eyeball scoring. The mitotic figure counts obtained through the automated algorithm and the pathologist’s manual score displayed a positive correlation and demonstrated a significant level of agreement. High mitotic scores derived from the three techniques were correlated with clinicopathological characteristics of aggressive tumour behaviour and were predictive of a poor patient survival (p<0.001); however, MAI was the only independent predictor of survival (p<0.05) in multivariate Cox regression analysis. The Ki 67 score positively correlated with MAI and MCT but not with MI.
Conclusions:
This thesis has highlighted the importance of mitosis-detection methods in BC. The implementation of DP and AI algorithms has shown promising results in improving the accuracy and reproducibility of mitosis scoring, overcoming the limitations of manual counting.
It also emphasised the need for standardised protocols and improved concordance in BC grading. Digital high-power screen fields on WSIs proved to be a practical approach for mitosis scoring, contributing to better grading and prognostic assessments.
Furthermore, the assessment of apoptosis alongside mitosis in BC has provided a more comprehensive understanding of tumour behaviour. The incorporation of apoptosis assessment into the grading system improved the accuracy of prognostic information and treatment decisions.
Novel staining techniques, such as the PHH3-H&E dual staining, offered a more specific and accurate identification of mitotic figures compared to traditional methods. This technique allowed for the simultaneous assessment of morphological features of mitosis and tumour histology, which led to improved diagnostic accuracy and prognostic value.
Moreover, the identification of specific genes associated with mitosis in BC provided insights into potential prognostic markers and therapeutic targets. These genes were found to be associated with poor clinicopathological parameters, increased proliferation, and shorter patient survival, which highlights their significance in BC biology and their potential for personalised medicine and improved outcomes.
The integration of DP, AI, standardised protocols, and novel staining techniques holds promise for improving mitosis scoring and overall patient care in BC pathology.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[thumbnail of Thesis_Asmaa Ibrahim (20144323).pdf]](https://eprints.nottingham.ac.uk/style/images/fileicons/application_pdf.png "Thesis_Asmaa Ibrahim (20144323).pdf")