Method for counting mitoses by image processing in Feulgen stained breast cancer sections

Accuracy and efficiency of an artificial intelligence tool when counting breast mitoses

BACKGROUND

The mitotic count in breast carcinoma is an important prognostic marker. Unfortunately substantial inter- and intra-laboratory variation exists when pathologists manually count mitotic figures. Artificial intelligence (AI) coupled with whole slide imaging offers a potential solution to this problem. The aim of this study was to accordingly critique an AI tool developed to quantify mitotic figures in whole slide images of invasive breast ductal carcinoma.

METHODS

A representative H&E slide from 320 breast invasive ductal carcinoma cases was scanned at 40x magnification. Ten expert pathologists from two academic medical centers labeled mitotic figures in whole slide images to train and validate an AI algorithm to detect and count mitoses. Thereafter, 24 readers of varying expertise were asked to count mitotic figures with and without AI support in 140 high-power fields derived from a separate dataset. Their accuracy and efficiency of performing these tasks were calculated and statistical comparisons performed.

RESULTS

For each experience level the accuracy, precision and sensitivity of counting mitoses by users improved with AI support. There were 21 readers (87.5%) that identified more mitoses using AI support and 13 reviewers (54.2%) that decreased the quantity of falsely flagged mitoses with AI. More time was spent on this task for most participants when not provided with AI support. AI assistance resulted in an overall time savings of 27.8%.

CONCLUSIONS

This study demonstrates that pathology end-users were more accurate and efficient at quantifying mitotic figures in digital images of invasive breast carcinoma with the aid of AI. Higher inter-pathologist agreement with AI assistance suggests that such algorithms can also help standardize practice. Not surprisingly, there is much enthusiasm in pathology regarding the prospect of using AI in routine practice to perform mundane tasks such as counting mitoses.

Analysis of images for detection of oral epithelial dysplasia: A review

This paper provides a review of various image analysis approaches that have been previously used for recognition of dysplasia in images of the epithelium of the oral cavity. This domain has become especially admissible with the uncovering of the importance of image analysis which can probably be an aid to subjective diagnosis by histopathologists. Oral malignancy is a rampant form of cancer found among people of the Indian subcontinent due to various deleterious habits like consumption of tobacco, areca nut, betel leaf etc. Oral Submucous Fibrosis, a precancer, whose pathological category falls between normal epithelium and epithelial dysplasia, is caused because of these habits and can ultimately lead to oral cancer. Hence early detection of this condition is necessary. Image analysis methods for this purpose have an enormous potential which can also reduce the heavy workload of pathologists and to refine the criterion of interpretation. This paper starts with a critique of statistics of oral carcinoma in India and distribution of cancer in intra-oral sites and moves on to its causes and diagnostic approaches including causative agents, problems in curative approach and importance of image analysis in cancer detection. The various image analysis methods to appraise the cytological and architectural changes accompanied by Oral Epithelial Dysplasia in the images of the oral epithelial region have been described in relation to 2005 WHO Classification System and it was found that in future, analysis of images based on the mentioned methods has the potential in better interpretation and diagnosis of oral carcinoma.

Image analysis assisted study of mitotic figures in oral epithelial dysplasia and squamous cell carcinoma using differential stains

BACKGROUND

Mitosis is a process of cell division resulting in two genetically equivalent daughter cells. Excessive proliferation of cells due to mitosis is the hallmark in pre cancer and cancer.

AIMS

This study was conducted to count the number of mitotic figures in normal oral mucosa, oral epithelial dysplasia and squamous cell carcinoma in both Hematoxylin and Eosin (H&E) and Crystal Violet stained sections. Also the overall number of mitotic figures with both stains were compared along with the evaluation of staining efficacy of both the stains.

METHODS AND MATERIAL

The present study was conducted on 20 specimens each of the three categories. These were further divided into two groups for staining with H&E and with 1% Crystal Violet respectively. Images were captured and analyzed using image analysis software Dewinter Biowizard 4.1.

RESULTS

Comparison of mitotic figure count in three categories in sections stained with both stains showed statistically significant difference (p < 0.001). The mean number of mitotic figures seen in Crystal Violet reagent were significantly higher as seen in H&E stain (p < 0.001). The overall diagnostic efficacy of Crystal Violet was 87.6%. Crystal Violet scored over H&E stain and also helped to better appreciate metaphases in Squamous cell carcinoma and telophases in dysplasia.

CONCLUSION

Number of mitotic figures progressively increase with the advancement of the pathology. Use of 1% Crystal Violet provides better appreciation of mitotic figures and can be employed as a selective stain in routine histopathology.

Breast cancer histopathology image analysis: a review

This paper presents an overview of methods that have been proposed for the analysis of breast cancer histopathology images. This research area has become particularly relevant with the advent of whole slide imaging (WSI) scanners, which can perform cost-effective and high-throughput histopathology slide digitization, and which aim at replacing the optical microscope as the primary tool used by pathologist. Breast cancer is the most prevalent form of cancers among women, and image analysis methods that target this disease have a huge potential to reduce the workload in a typical pathology lab and to improve the quality of the interpretation. This paper is meant as an introduction for nonexperts. It starts with an overview of the tissue preparation, staining and slide digitization processes followed by a discussion of the different image processing techniques and applications, ranging from analysis of tissue staining to computer-aided diagnosis, and prognosis of breast cancer patients.

Assessment of algorithms for mitosis detection in breast cancer histopathology images

The proliferative activity of breast tumors, which is routinely estimated by counting of mitotic figures in hematoxylin and eosin stained histology sections, is considered to be one of the most important prognostic markers. However, mitosis counting is laborious, subjective and may suffer from low inter-observer agreement. With the wider acceptance of whole slide images in pathology labs, automatic image analysis has been proposed as a potential solution for these issues. In this paper, the results from the Assessment of Mitosis Detection Algorithms 2013 (AMIDA13) challenge are described. The challenge was based on a data set consisting of 12 training and 11 testing subjects, with more than one thousand annotated mitotic figures by multiple observers. Short descriptions and results from the evaluation of eleven methods are presented. The top performing method has an error rate that is comparable to the inter-observer agreement among pathologists.

Image processing of hematoxylin and eosin-stained tissues for pathological evaluation

Color and geometric characteristics of stained areas in histochemical slides are among the features pathologists assess to evaluate the severity of lesions. In this research, image processing techniques were used to perform objective quantification of these characteristics in images of H&E-stained spleen tissues. A segmentation algorithm was developed to isolate the areas of interest in microscopic tissue images. Image features important to pathological evaluation were then extracted. These features were used to build statistical and neural network models to predict pathologist scores. A linear regression model predicted the scores to an R(2)-value of 0.6, and a neural network model classified samples to an accuracy of 75%. The results show the usefulness of image processing as a tool for pathological evaluation.

Gross genomic damage measured by DNA image cytometry independently predicts gastric cancer patient survival

Background:

DNA aneuploidy reflects gross genomic changes. It can be measured by flow cytometry (FCM-DNA) or image cytometry (ICM-DNA). In gastric cancer, the prevalence of DNA aneuploidy has been reported to range from 27 to 100%, with conflicting associations with clinicopathological variables. The aim of our study was to compare the DNA ploidy status measured using FCM-DNA and ICM-DNA in gastric cancer and to evaluate its association with clinicopathological variables.

Methods:

Cell nuclei were isolated from 221 formalin-fixed, paraffin-embedded gastric cancer samples. DNA ploidy was assessed using FCM-DNA and ICM-DNA.

Results:

A total of 178 (80.5%) gastric cancer samples were classified as DNA aneuploid using FCM-DNA, compared with 172 (77.8%) gastric cancer samples when using ICM-DNA. Results obtained from both methods were concordant in 183 (82.8%) cases (κ=0.48). Patients with ICM-DNA diploid gastric cancer survived significantly longer than those with ICM-DNA aneuploid gastric cancer (log rank 10.1, P=0.001). For FCM-DNA data, this difference did not reach statistical significance. The multivariate Cox model showed that ICM-DNA ploidy status predicted patient survival independently of tumour-node-metastasis status.

Conclusion:

ICM-DNA ploidy status is an independent predictor of survival in gastric cancer patients and may therefore be a more clinically relevant read out of gross genomic damage than FCM-DNA.

Detecting mitoses in time-lapse images of embryonic epithelia using intensity analysis

Although the frequency and orientation of mitoses can significantly affect the mechanics of early embryo development, these data have not been available due to a shortage of suitable automated techniques. Fluorescence imaging, though popular, requires biochemical intervention and is not always possible or desirable. Here, a new technique that takes advantage of a localized intensity change that occurs in bright field images is used to identify mitoses. The algorithm involves mapping a deformable, sub-cellular triangular mesh from one time-lapse image to the next so that corresponding regions can be identified. Triangles in the mesh that undergo darkening of a sufficient degree over a period consistent with mitosis are flagged. Mitoses are assumed to occur along the short axis of elliptical areas fit to suitably sized clusters of flagged triangles. The algorithm is less complex than previous approaches and it has strong discrimination characteristics. When applied to 15 image sets from neurulation-stage axolotl (Ambystoma mexicanum) embryos, it was able to correctly detect 86% of the manually identified mitoses, had less than 5% false positives and produced average angular errors of only 15 degrees . The new algorithm is simpler to implement than those previously available, is substantially more accurate, and provides data that is important for understanding the mechanics of morphogenetic movements.

Automatic breast cancer grading of histopathological images

Breast cancer grading of histopathological images is the standard clinical practice for the diagnosis and prognosis of breast cancer development. In a large hospital, a pathologist typically handles 100 grading cases per day, each consisting of about 2000 image frames. It is, therefore, a very tedious and time-consuming task. This paper proposes a method for automatic computer grading to assist pathologists by providing second opinions and reducing their workload. It combines the three criteria in the Nottingham scoring system using a multi-resolution approach. To our best knowledge, there is no existing work that provide complete grading according to the Nottingham criteria.

A survey on histological image analysis-based assessment of three major biological factors influencing radiotherapy: proliferation, hypoxia and vasculature

Image analysis is a rapidly evolving field with growing applications in science and engineering. In cancer research, it has played a key role in advancing techniques of major diagnostic importance, minimising human intervention and providing vital clinical information. Especially in the field of tissue microscopy, the use of computers for the automated analysis of histological sections is becoming increasingly important. This paper presents an overview of various image analysis methodologies and summarises developments in this field, with great emphasis given on the assessment of three major biological factors known to influence the outcome of radiotherapy: proliferation, vasculature and hypoxia. A brief introduction followed by a survey is provided in each of these areas.

DNA cytometric features in biopsies of TaT1 urothelial cell cancer predict recurrence and stage progression more accurately than stage, grade, or treatment modality

OBJECTIVES

To compare retrospectively the predictive value for recurrence and stage progression of DNA ploidy and S-phase fraction by flow cytometry and highly automated ultrafast image cytometry (ICM) in biopsies of TaT1 urothelial cell carcinomas (UCCs) of the urinary bladder with stage, grade, other pathologic features, and treatment.

METHODS

Three experienced pathologists reviewed the stage and grade of 228 UCCs; 193 (85%) consensus cases were analyzed further. We had enough material for single-cell suspensions for both flow cytometry and ICM in 183 cases (94.8%). The 2001 European Society for Analytical Cellular Pathology standards for DNA ICM were followed. The predictive value of DNA features, classic prognosticators (stage, grade, carcinoma in situ, multicentricity), and treatment modality for recurrence and stage progression were analyzed with univariate (Kaplan-Meier) survival and multivariate (Cox model) regression analysis. Ta and T1 cases were analyzed separately.

RESULTS

Of the 228 cases, 88 (51.5%) recurred and 13 (7.6%) progressed. On univariate analysis, most of the DNA features studied were statistically significant. Treatment modality and grade were only prognostic for progression (not for recurrence) and only in Ta cases. On multivariate analysis, DNA ICM features performed best; the strongest recurrence predictor for Ta UCC was a DNA index (DI) of 1.0 versus all others, and for T1 UCC, a DI of less than 1.3 versus 1.3 or greater. The best stage progression predictor for Ta UCCs was a DI of 1.0 plus an S-phase fraction of less than 10%, and for T1 UCCs, a DI of less than 1.3 versus 1.3 or greater. With multivariate analysis, sex, age, grade, carcinoma in situ, multicentricity, and treatment modality were excluded once the DNA ICM features were selected.

CONCLUSIONS

DNA image cytometric features predict recurrence and stage progression in TaT1 UCC biopsies more accurately than classic prognostic factors, independent of treatment modality.

A histopathological contribution to supratentorial glioma grading, definition of mixed gliomas and recognition of low grade glioma with Rosenthal fibers

BACKGROUND

Previous glioma studies have described separate grading systems for oligodendrogliomas and astrocytomas. Many of these gliomas contain mixtures of neoplastic astrocytes and oligodendrocytes. Prognosis may be related to the percentages of these neoplastic components. Previous survival/grading studies have been limited to histopathological features but have not evaluated the importance of percentages of neoplastic components. This study attempted to perceive the relative importance of percentages of neoplastic astrocytes and oligodendrocytes for definition of astroglial, oligodendroglial and mixed oligoastroglial tumors. After determination of these limits we explored the possibility to develop a grading system for common supratentorial gliomas based on reproducible histopathological features.

METHODS

A retrospective study was performed of 362 cases of unselected supratentorial glioma. One hundred and thirty-eight binary and nine continuous histopathological variables, amongst which percentages of neoplastic astrocytes and oligodendrocytes, were scored and related to survival. Only well reproducible histological features were accepted in Cox regression to define glioma grades.

RESULTS AND CONCLUSIONS

Supratentorial gliomas appeared to be composed of variable percentages of neoplastic oligodendrocytes and astrocytes, but this spectrum did not correspond to a continuous change in prognosis. Gliomas containing 30% or more neoplastic oligodendrocytes had a slightly better outcome (p < 0.0432) but higher percentages did not further improve prognosis. Percentages of neoplastic astrocytes were not correlated to survival. We therefore propose to designate gliomas containing 30% or more neoplastic oligodendrocytes as oligodendroglial tumors, and others as astroglial tumors. From a prognostic point of view there is no need to recognize mixed oligoastrocytomas. An interesting finding was the recognition of a low grade glioma group with Rosenthal fibers, which had the longest postoperative survival. Another prognosticator of interest concerns the mitotic rate as a continuous variable. Atypical mitoses indicated the worst survival, after necrosis. It was possible to develop a grading system for all supratentorial gliomas using six reproducible histological parameters: necrosis, atypical mitoses, the mitotic rate, endothelial proliferative activity, percentage of neoplastic oligodendrocytes and Rosenthal fibers. This resulted in four grades for astroglial tumors (p < 0.002) and three grades for oligodendroglial tumors (p < 0.008) which differed significantly within each group with respect to survival.

Relationships between vascularization and proliferation in invasive breast cancer

Studies on relationships between angiogenesis and tumour cell proliferation have provided conflicting results. This study has therefore investigated the relationships between the number and location of fully automatically identified CD31-positive microvessels and interactively segmented mitoses and necrotic compartments by image processing. These features were studied in ten invasive breast cancers, in the 'hot spots' and in whole tumour sections. Microvessel and mitosis hot spots were topographically close or overlapping and were always located at the periphery of the tumour. The numbers of mitoses and microvessels per mm(2) in the hot spot were strongly correlated with the respective numbers in the whole tumour section, as well as mutually. The ratio of mitoses in the hot spot to the whole tumour section was significantly higher than the corresponding microvessel ratio. Mitoses were preferentially located at a distance of 50-150 microm from microvessels. No significant difference was found between the average distance between mitoses and microvessels in the whole tumour sections and in the hot spot (79 vs. 72 microm), although considerable inter-tumour differences were found (hot spot 43-101 microm, tumour 47-111 microm). The presence of necrotic areas correlated with the number of mitoses per mm(2) and necrosis was in general observed at a distance of more than 150 microm from the microvessels, suggesting that necrotic areas have outgrown their vascular system. These results indicate the usefulness of image processing of whole tumour sections for the identification of proliferation and vascularization hot spots, which are strong prognostic factors in breast cancer. The results also support a close relationship between tumour necrosis and microvessels.

Fully automated microvessel counting and hot spot selection by image processing of whole tumour sections in invasive breast cancer

BACKGROUND

Manual counting of microvessels is subjective and may lead to unacceptable interobserver variability, which may explain conflicting results.

AIMS

To develop and test an automated method for microvessel counting and objective selection of the hot spot, based on image processing of whole sections, and to compare this with manual selection of a hot spot and counting of microvessels.

METHODS

Microvessels were stained by CD31 immunohistochemistry in 10 cases of invasive breast cancer. The number of microvessels was counted manually in a subjectively selected hot spot, and also in the same complete tumour sections by interactive and automated image processing methods. An algorithm identified the hot spots from microvessel maps of the whole tumour section.

RESULTS

No significant difference in manual microvessel counts was found between two observers within the same hot spot, and counts were significantly correlated. However, when the hot spot was reselected, significantly different results were found between repeated counts by the same observer. Counting all microvessels manually within the entire tumour section resulted in significantly different hot spots than manual counts in selected hot spots by the same observer. Within the entire tumour section no significant differences were found between the hot spots of the manual and automated methods using an automated microscope. The hot spot was found using an eight connective path search algorithm, was located at or near the border of the tumour, and (depending on the size of the hot spot) did not always contain the field with the largest number of microvessels.

CONCLUSIONS

The automated counting of microvessels is preferable to the manual method because of the reduction in measurement time when the complete tumour is scanned, the greater accuracy and objectivity of hot spot selection, and the possibility of visual inspection and relocation of each measurement field afterwards.

Evaluating mitotic activity in canine and feline solid tumors: standardizing the parameter

Three different methods for evaluating mitotic activity (mitotic count, mitoses/area, mitotic index) were applied to different types of canine and feline solid tumors to determine the method that is most objective and correlates best with other parameters of cell proliferation. Mitotic activity was evaluated on toluidine blue stained histological sections. Slides stained with histochemical (AgNOR proteins) and immunohistochemical (MIB1, PCNA) markers of cell proliferation were available for each case. Quantitation of mitotic activity and cell proliferation parameters was performed with an image analyzer. Mitotic activity assessment was compared with cell proliferation indices and its ability to discriminate tumors grouped on histologically based criteria including the histological type, malignant or benign characteristics, and grade. A significant correlation by linear regression analysis with other parameters assessing cell proliferation revealed that mitotic index correlated 1000% and mitoses/area and mitotic count correlated 40% of the time. In discriminating the proliferative activity of tumors grouped by histological criteria, mitotic index and mitotic count revealed 1000% concordance with the other parameters of cell proliferation, while mitoses/areas showed 80% concordance.

Origins of ... image analysis in clinical pathology

Images

An evaluation of prognostic factors in advanced ovarian cancer

A summary is presented of currently available prognostic factors in advanced ovarian cancer of the common epithelial types. The emphasis is on the most promising clinical, classical pathological, biochemical, immunohistochemical, molecular biological and quantitative pathological factors.

Comparison of the prognostic value of four methods to assess mitotic activity in 186 invasive breast cancer patients: classical and random mitotic activity assessments with correction for volume percentage of epithelium

Proliferation markers and especially the Mitotic Activity Index (MAI) are strong and reproducible prognosticators in invasive breast cancer. Traditionally, the MAI has been defined as the total number of mitoses counted in 10 consecutive high-power fields (objective, x40; numeric aperture, .75; field diameter, 450 microns), in the most cellular area at the periphery of the tumor, with the subjectively highest mitotic activity. No correction for epithelial percentage or cellularity was applied. This study investigates whether the prognostic value of mitotic activity could be improved by a random sampling procedure or correction for percentage of epithelium present. For this purpose the prognostic value of four methods used to assess mitotic activity in invasive breast cancer was compared in 4-microns-thick hematoxylin-eosin (H&E)-stained sections of 186 primary invasive breast cancer patients. These were the MAI, the random MAI (rMAI), the Mitosis per Volume (M/V) Index, and the random M/V Index (rM/V Index). The rMAI was defined as the total number of mitotic figures counted in 10 random fields through the whole outlined tumor at x400 magnification. A correction for the volume percentage of epithelium assessed with stereology yielded the M/V Index and the rM/V Index, respectively. The results of all four methods showed moderate to high correlations. Univariate survival analysis (Kaplan-Meier curves; Mantel-Cox test) confirmed that all four methods had a strong prognostic value (P < .001). The MAI, however, produced the best results (Mantel-Cox value, 17.1). Multivariate analysis showed that all four methods had additional prognostic value to tumor size and lymph node status. The M/V Index provided most additional prognostic information, followed by the MAI. Assessment of rMAI took 20 to 30 minutes on average, about two times longer than MAI. The correction for volume percentage of epithelium took about 10 minutes longer for both methods than the uncorrected methods. In conclusion, the rMAI gives an impression of the mitotic activity through the whole tumor, with almost similar prognostic value as the traditional MAI, especially when correcting for percentage of epithelium. Nevertheless, the MAI is still to be preferred, because the assessment is easy to apply and less time consuming.

Tumor vascularization, mitotic index, histopathologic grade, and DNA ploidy in the assessment of 114 head and neck squamous cell carcinomas

BACKGROUND

Quantification of tumor vascularization recently has been shown to a parameter of potential clinical significance. Several basic and clinical studies have demonstrated that tumor growth correlates significantly with angiogenesis.

METHODS

To determine the utility of quantification of tumor vascularization and mitotic index for the pathobiologic assessment of head and neck squamous cell carcinoma, a prospective study of 114 consecutively recruited primary neoplasms was performed. Tumors were also studied for differentiation, keratinization, nuclear atypia, growth pattern, inflammation, desmoplasia, vascular tumor emboli, and DNA content.

RESULTS

In this cohort, tumor vascularization was correlated with mitotic index (P < 0.001), nuclear grade (P = 0.03), presence of tumor emboli in the peripheral microvessels (P = 0.05), and lymph nodal status (P = 0.03). A strong relationship between poor differentiation and high N classification (P < 0.001), differentiation and keratinization (P < 0.001) and tumor cell emboli and clinically involved lymph nodes (P = 0.01) was also observed. Emboli were more rare in laryngeal and oropharynx/oral cavity tumors than in hypopharynx/epilarynx (P = 0.02).

CONCLUSIONS

This study indicates that tumor vascularization, differentiation, and tumor emboli in peripheral microvessel network are important histologic parameters in the assessment of squamous cell carcinoma of the head and neck.

Quantitative microscopical and confocal laser scanning microscopy for intermediate endpoint biomarkers in breast cancer: potential and reproducibility

Diagnostic quantitative pathological (QP) determinations are increasingly used in our hospital. The number of requests for QP for reference materials is rising rapidly. This is understandable; quantitative assessments have a strong prognostic value and can be very reproducible, depending on the care taken with a number of factors including cell and tissue processing, application of the appropriate stains, and the measurement protocol used. As to the latter, systematic random sampling gives the best intra- and interobserver agreement (with correlation coefficients between observers for certain features > or = 0.94). Flow cytometric determinations are often regarded as more reproducible than interactive morphometry due to the high speed of the assessments, the large number of objects measured per specimen, and the lack of observer interaction. Indeed, flow cytometrically assessed DNA ploidy is very reproducible, even though the % S-phase fraction is much more variable. Unlike image cytometry (ICM), visual inspection of cells is not easily accomplished with flow cytometry (FCM). With ICM, the fully automated measurement of DNA in thousands of cells is possible in 3-5 minutes, with a very low coefficient of variation (< or = 2% for the diploid and tetraploid peak of liver cell nuclei). ICM also allows measurement of texture features. However, quantitative immunohisto/cytochemical determinations may not always be as reproducible as sometimes believed. Recently, we found large variations in the measurements, made by a commercially available image processing instrument, of the estrogen and progesterone receptors, Ki-67, cathepsin D, and neu protein overexpression in breast cancer.(ABSTRACT TRUNCATED AT 250 WORDS)

The Multi-Center Morphometric Mammary Carcinoma Project (MMMCP) in The Netherlands: value of morphometrically assessed proliferation and differentiation

The Multi-Center Morphometric Mammary Carcinoma Project (MMMCP) was set up to investigate the reproducibility and prognostic value of routine assessments of morphometric parameters [mean nuclear area (MNA), mitotic activity index (MAI), and multivariate prognostic index (MPI)] and cytometric features (DNA ploidy and index, % S-phase cells, as well as other cell cycle data) in comparison with classical prognostic parameters and steroid receptors. Thirty-four hospitals in six geographic regions participated. In 1988-1989, 3427 patients entered the study and morphometric assessments were made. An interim (1993) survival analysis indicated that MAI, MNA, and MPI are the strongest predictors of outcome. A Phase III randomized prospective multi-center trial [Premenopausal Morphometric Intervention Study (PREMIS)] using these endpoints was initiated in Europe to evaluate adjuvant [cyclophosphamide, methotrexate and 5-fluorouracil (CMF)] chemotherapy versus observation in morphometrically high risk (i.e., MAI > 10), premenopausal, lymph node negative (LN-) breast cancer patients.