More about BCDR
While breast cancer databases have been reported up to date for studying breast cancer, the information included in these databases often presents some undesirable issues: a) some are incomplete in terms of available features (image-based descriptors, clinical data, etc.); b) others have a small number of annotated patients’ cases; c) and/or the database is private and cannot be used as reference, which makes difficult exploring and comparing the performance of computer-based detection and diagnosis methods.
This research aims to develop a wide-ranging annotated BREAST CANCER DIGITAL REPOSITORY (BCDR) with two main objectives: (1) establishing a novel reference to explore computer-based detection and diagnosis methods, and (2) training medical students, formed physicians and other medical-related professionals. The creation of BCDR was supported by the IMED Project (for Development of Algorithms for Medical Image Analysis) aimed at creating medical image repositories and massive exploration of Computer-Aided Diagnosis (CADx) methods on GRID computing resources. The IMED project was carried out by INEGI, FMUP-CHSJ – University of Porto, Portugal and CETA-CIEMAT, Spain between March 2009 and March 2013. Recently, in October of 2013 the IMED project was renewed and Aveiro University began to be part of this consortium. Now, the four institutions continue to actively augment and develop the BCDR.
Currently, the BCDR contains cases of 1734 patients with mammography and ultrasound images, clinical history, lesion segmentation and selected pre-computed image-based descriptors. Patient cases are BIRADS classified and annotated by specialized radiologists.
The BCDR is subdivided in two different repositories: (1) a Film Mammography-based Repository (BCDR-FM) and (2) a Full Field Digital Mammography-based Repository (BCDR-DM). Both repositories were created with anonymous cases from medical archives (complying with current privacy regulations as they are also used to teach regular and postgraduate medical students) supplied by the Faculty of Medicine – Centro Hospitalar São João, at University of Porto (FMUP-HSJ). BCDR provides normal and annotated patients cases of breast cancer including mammography lesions outlines, anomalies observed by radiologists, pre-computed image-based descriptors as well as related clinical data.
The BCDR-FM is composed by 1010 (998 female and 12 male) patients cases (with ages between 20 and 90 years old), including 1125 studies, 3703 mediolateral oblique (MLO) and craniocaudal (CC) mammography incidences and 1044 identified lesions clinically described (820 already identified in MLO and/or CC views). With this, 1517 segmentations were manually made and BI-RADS classified by specialized radiologists. MLO and CC images are grey-level digitized mammograms with a resolution of 720 (width) by 1168 (height) pixels and a bit depth of 8 bits per pixel, saved in the TIFF format.
The BCDR-DM, still in construction, at the time of writing is composed by 724 (723 female and 1 male) Portuguese patients cases (with ages between 27 and 92 years old), including 1042 studies, 3612 MLO and/or CC mammography incidences and 452 lesions clinically described (already identified in MLO and CC views). With this, 818 segmentations were manually made and BI-RADS classified by specialized radiologists. The MLO and CC images are grey-level mammograms with a resolution of 3328 (width) by 4084 (height) or 2560 (width) by 3328 (height) pixels, depending on the compression plate used in the acquisition (according to the breast size of the patient). The bit depth is 14 bits per pixel and the images are saved in the TIFF format.
The BCDR was released for public domain in April, 18, 2012 and it is still in development. Currently, four benchmarking datasets (two masses-based and two microcalcifications and calcifications-based) representatives of benign and malignant lesions (biopsy-proven) comprising instances of clinical and image-based features are available for free download to registered users. In the near future it is our intention to continue publishing new benchmarking datasets.
