CADMAMMO : “Innovative Computer-Aided Decision making system, combining MAMMOgraphic, histological and cytological image data, for improving breast cancer clinical management”

Cancer is a major cause of morbidity and mortality in Europe. In 2006, 3.191.600 new cases of cancer were diagnosed and 1.703.000 deaths from cancer were registered. Breast cancer is the most frequent malignant tumour among women and the fifth most common cause of cancer death. Diagnostic mammography is the most important and reliable screening method for confirming the location, extent, and other important clinical features of breast tumours. Mammograms are interpreted in accordance with the BI-RADS (Breast Imaging Reporting and Database System) system for classifying mammograms, with regards to the absence or likelihood of breast cancer. Among the most important signs indicating the presence of malignant lesions is the existence of masses, microcalcifications (mCs) and clusters of mCs. However, in early stage cancer, the subtle differences between normal and abnormal tissues have been proven challenging for viewing even by experienced physicians: up to 30% of breast lesions are missed during routine diagnosis. This is one of the reasons why mammograms are considered to be among the most difficult to interpret types of medical images. Another important reason is the low contrast appearance of early stage abnormalities (such as mCs) and their low differentiation from the surrounding breast tissues. Microcalcifications should appear as bright spots, however, in many cases, due to the composition of surrounding breast tissues and imaging limitations, microcalcifications appear as low contrast entities. Even though mammography has been dramatically improved over the past years, in many cases of breast lesion(s) diagnosis may be uncertain. Additionally, although imaging findings may indicate the existence of an abnormality, however, the actual diagnosis might not be obvious. Even when an abnormality is obvious, radiological findings cannot be solely used for predicting the course of the disease or for specifying of the treatment planning. In the presence of suspected lesion(s), pathologists are called to reach the final diagnostic conclusion on the basis of visual evaluation of microscopic material (histological and/or cytological) under the microscope for the clarification of the importance of certain biological factors, such as histological tumour grade and estrogens reception (ER) status. Grading is determined on the visual estimation of certain histological features, such as glandular differentiation, nuclear pleomorphism, and mitotic count, on H&E (Hematoxylin & Eosin) stained specimens according to the World Health Organization (WHO) guidelines, whereas ER-status is assessed as the percentage of expressed nuclei on immunohistochemically stained (IHC) specimens as suggested by the American Society of Clinical Oncology (ASCO) protocol. Although microscopic examination is a critical process for treatment planning, the potential of diagnostic errors still remains substantially high; even the most experienced pathologist’s diagnosis is sometimes a biased opinion, based on experience and on loosely defined classification criteria. Factors that affect diagnostic accuracy include experts’ subjectivity and lack of experience, inter observer diagnostic variability, tumours’ heterogeneity – (the higher the degree of malignancy the more heterogeneous is the composition of the tumour, the fact that tumours develop along a morphological and biological continuum, and poor sampling of tumour’s tissues) so that not always the most representative tissue region is selected for examination.

Recent literature has highlighted that diagnostic, prognostic, and predictive misinterpretations in breast cancer management can be reduced by combining information from both radiological and microscopy findings. However, to the best of our knowledge, no attempt for quantification, analysis, and computer-based interpretation of the integrated information from both mammography and tissue/cytological imaging has been reported in literature.

The aim of CADMAMMO is to design, develop and implement in the clinical routine a hybrid multi-modal (radiological and microscopic imaging) computer-aided diagnostic tool for improving diagnostic, prognostic, and predictive classification of patients’s breast cancer tumors. Thus, CADMAMMO will lead to:

  • More accurate diagnostic, prognostic, and predictive assessments leading to more precise treatment planning
  • Reduction of the potential of diagnostic misinterpretations through more accurate determination of the type (benign/malignant), the degree of malignancy, and the level of ER-receptors’ status
  • Production of new scientific knowledge, by investigating the correlation between radiological findings (mammography) with histological and cytological imprints of suspected lesions
  • Design and implementation of a dynamic, web-based decision support system on a Graphics Processing Unit (GPU) framework, able to adjust its design based on the exact clinical question to be answered (for example classification based on the BI-RADS or the WHO system) and update its structure in real time whenever a new verified case is uploaded on its repository.

CADMAMMO will be the first coordinated response towards integration of information from radiological and microscopy imaging for improving diagnostic, prognostic, and predictive efficiency in breast cancer management. CADMAMMO will make progress beyond state-of-art in the following issues:

  1. CADMAMMO will introduce an innovative computer-aided decision making strategy combining information from mammography, histology, and cytology; new findings are expected to emerge by investigating through the CADMAMMO platform the correlation between radiological findings with biological expressions on the tissue level, such as degree of malignancy and ER-status. To the best of our knowledge, the latter correlation is for the first time investigated,
  2. Considering that that the training, experience, and expertise of the radiologist and pathologist in interpreting breast imaging examinations are of paramount importance, the contribution of the operator-independent strategies that CADMAMMO introduces will enable objective and reliable estimations of crucial imaging findings, such as presence/absence of microcalcifications and ER-status of suspected lesions,
  3. Diagnostic, prognostic and predictive decisions based on the integrated multimodal information will be linked to treatment planning for more accurate selection of therapeutic strategies,
  4. CADMAMMO will investigate methods for generating classifier ensemble schemes based on stochastic formulation, layered-clustering and hierarchical analysis, in order to specify the diagnostic, prognostic and/or predictive importance of each feature (radiological, cytological, and histological) in the final classification result,
  5. The decision support system will be implemented in a dynamic web-based platform, built on a Graphics Processing Unit (GPU) framework in order for the decision support system to be able to update its structure in real time whenever a new verified case is uploaded on its repository and to be able to adjust its design based on the exact clinical question to be answered (for example classification based on the BI-RADS or the WHO system),
  6. The platform will be designed for exploitation even from remote medical sites without tissue-blocks, or tissue slides having to be moved from local departments to specialized centers for second readings,
  7. The platform will be designed to be upgradeable to more willing to join partners.

STATUS: In Progress



MEDICAL IMAGE & SIGNAL PROCESSING Lab. Department of Biomedical Engineering – Τ.Ε.Ι. of Athens

Menu Title