Abstract
IntroductionPathologists currently diagnose breast lesions through histologic assessment, which requires fixation and tissue preparation. The diagnostic criteria used to classify breast lesions are qualitative and subjective, and inter-observer discordance has been shown to be a significant challenge in the diagnosis of selected breast lesions, particularly for borderline proliferative lesions. Thus, there is an opportunity to develop tools to rapidly visualize and quantitatively interpret breast tissue morphology for a variety of clinical applications.MethodsToward this end, we acquired images of freshly excised breast tissue specimens from a total of 34 patients using confocal fluorescence microscopy and proflavine as a topical stain. We developed computerized algorithms to segment and quantify nuclear and ductal parameters that characterize breast architectural features. A total of 33 parameters were evaluated and used as input to develop a decision tree model to classify benign and malignant breast tissue. Benign features were classified in tissue specimens acquired from 30 patients and malignant features were classified in specimens from 22 patients.ResultsThe decision tree model that achieved the highest accuracy for distinguishing between benign and malignant breast features used the following parameters: standard deviation of inter-nuclear distance and number of duct lumens. The model achieved 81 % sensitivity and 93 % specificity, corresponding to an area under the curve of 0.93 and an overall accuracy of 90 %. The model classified IDC and DCIS with 92 % and 96 % accuracy, respectively. The cross-validated model achieved 75 % sensitivity and 93 % specificity and an overall accuracy of 88 %.ConclusionsThese results suggest that proflavine staining and confocal fluorescence microscopy combined with image analysis strategies to segment morphological features could potentially be used to quantitatively diagnose freshly obtained breast tissue at the point of care without the need for tissue preparation.Electronic supplementary materialThe online version of this article (doi:10.1186/s13058-015-0617-9) contains supplementary material, which is available to authorized users.
Highlights
Pathologists currently diagnose breast lesions through histologic assessment, which requires fixation and tissue preparation
Fixed tissue samples are processed after harvesting and are evaluated for presence and type of malignant breast tissue based on standardized histologic criteria [1,2,3,4], which employ cytological and qualitative architectural features
We show that combining both yields improved diagnostic performance, in the diagnosis of invasive ductal cancer (IDC) and ductal carcinoma in situ (DCIS) The nuclear and ductal parameters described in this study could potentially be used for objective categorization of breast lesions
Summary
Pathologists currently diagnose breast lesions through histologic assessment, which requires fixation and tissue preparation. Dobbs et al Breast Cancer Research (2015) 17:105 of techniques that use quantitative criteria that can be applied without subjecting the tissue to processing can overcome the subjectivity of interpretation and may reduce the inter- and intra-observer variability in the histological evaluation of breast tissue [12]. Such techniques could be potentially useful in settings lacking the human resources and equipment necessary to perform standard histologic assessment, which can be a challenge in many parts of the world [13]. Micro-optical computed tomography and nuclear morphometry was used to compare variations between human breast cell lines and found that nuclear volumes increased from normal to metastatic breast cells and that nuclei of abnormal cells contained more nucleoli [23]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have