BUSAT: Breast Ultrasound Analysis Toolbox

The Breast Ultrasound Analysis Toolbox contains 70 functions (m-files) to perform image analysis including: image preprocessing, lesion segmentation, morphological and texture features, and binary classification (commonly benign and malignant classes). The implementations were developed in Matlab R2014a.

Any use of this toolbox please reference our article: Arturo Rodríguez-Cristerna, Wilfrido Gómez-Flores, and Wagner Coelho de Albuquerque-Pereira, ''BUSAT: A MATLAB Toolbox for Breast Ultrasound Image Analysis'', In: 9th Mexican Congress on Pattern Recognition (MCPR 2017), LNCS 10267, pp. 268-277, 2017. DOI: 10.1007/978-3-319-59226-8_26.

New functions:

  • bootstrap632AUC: AUC estimation using the .632+ bootstrap method.
  • bootstrap632ERR: Classification error estimation using the .632+ bootstrap method.
  • trainRF: Train random forest classifier.
  • classifyRF: Classify data with random forest classifier.
  • phasecong: Phase congruency of a gray-scale image.
  • lbpv: Local binary pattern variance from phase congruency.

  • DOWNLOAD CODES (ZIP-3.7 Mb) updated on June 20, 2017.
  • The authors thank users for helpful discussion and comments. Please send comments and suggestions to .

    Image segmentation with evolutionary PCNN

    The pulse-coupled neural network (PCNN) is based on the cortical model proposed by Eckhorn and is widely used in tasks such as image segmentation. The PCNN performance is particularly limited by adjusting its input parameters, where computational intelligence techniques have been used to solve the problem of PCNN tuning. However, most of these techniques use the entropy measure as a cost function, regardless of the relationship of inter-/intra-group dispersion of the pixels related to the objects of interest and their background. Therefore, we propose using the differential evolution algorithm along with a cluster validity index as a cost function to quantify the segmentation quality in order to guide the search to the best PCNN parameters to get a proper segmentation of the input image. The numerical implementation of was developed in Matlab R2014a.

    Any use of this program please reference the article Automatic tuning of the pulse-coupled neural network using differential evolution for image segmentation, in Mexican Congress on Pattern Recognition 2016, LNCS 9703, pp. 157-166, 2016.

  • 3D segmentation of X-ray μCT

    The quantification of trabecular bone morphology is done by histomorphometric analysis, which is based on stereologic principles, that is, by inferring 3D information from planar sections. To compute the bone histomorphology, it is necessary to extract the volume of interest (VOI) from the 3D data, which means separating the trabecular bone tissue from the background. Because x-ray microtomography (μCT) consists of a number of 2D grayscale digital frames, thresholding techniques are useful to extract the trabecular bone volume. Four automatic thresholding algorithms were implemented: clustering, maximum entropy, moment preservation, and concavity-based. The numerical implementation of such methods were developed in Matlab 2012a.

    Any use of this program please reference the article A comparative study of automatic thresholding approaches for 3D X-ray microtomography of trabecular bone, published in Medical Physics, vol. 40, no. 9, pp. 091903-1-14, 2013.

  • Note: this program was tested under Mac OS X (10.6.8) with Matlab 2012a.
  • Some UCI datasets ready for MATLAB

    There are provided 31 common datasets used to evaluate classifiers. These datasets were downloaded from the UCI Machine Learning Repository. A single dataset saved in a MAT file could be loaded to MATLAB environment by using the 'load' function. The variable 'X' is the attribute matrix of size NxD (instances by attributes). The variable 'Y' is the class label vector of size Nx1 with values {1,2,...,c }. The datasets do not have missing values.

    Dataset Attributes Classes Instances
    Breast Cancer Wisconsin Original 9 2 683
    Breast Cancer Wisconsin Diagnosis 30 2 569
    Breast Cancer Wisconsin Prognostic 33 2 194
    Car Evaluation 6 4 1728
    Cardiotocography 34 3 & 10 2126
    Climate Model Simulation Crashes 18 2 540
    Credit Approval 15 2 653
    Dermatology 34 6 358
    Diabetic Retinopathy Debrecen 19 2 1151
    Echocardiogram Data 9 2 61
    Glass Identification 9 6 214
    Heart Disease Cleveland 13 2 297
    Hepatitis Domain 18 2 112
    Indian Liver Patient Dataset 10 2 579
    Ionosphere 33 2 351
    Iris data 4 3 150
    Leaf 14 30 340
    BUPA Liver Disorders 6 2 341
    Mice Protein Expression 80 8 552
    Ozone Level Detection 72 2 1848
    Parkinsons Data Set 22 2 195
    Pima Indians Diabetes 8 2 768
    Seeds 7 3 199
    SPECTF Heart Data Set 44 2 267
    Statlog (Image Segmentation) 18 7 2310
    Steel Plates Faults 27 7 1941
    Thyroid gland data 5 3 215
    US Congressional Voting Records 16 2 232
    Vehicle Silhouettes 18 4 846
    Vowel Recognition Data 11 11 990
    Wine recognition data 13 3 178

  • Download all datasets (ZIP-1.4 Mb)