2. Development of a CADx system to identify breast abnormalities in digital mammograms images using artificial neural networks

In this research, the study of BC disease using advanced techniques of DIP, KDD, and AI was carried out in order to develop imaging biomarkers that allow to carry out diverse studies for BCD. As is showed in Figure 1, the research was divided in three main stages.

Currently, there are no public databases of BC in Latin America or Mexico. Therefore, at first stage, different public mammography databases were used for developing and validating digital image processing algorithms capable to select ROIs from mammograms to extract


Figure 1. Main stages of research.

Due to all the problems presented by mammography screening, great efforts have been made to support the radiologist in the search for these lesions through computer-aided detection (CAD) or biomarker systems, which try to help the radiologist by taking advantage of the

At present, computer-aided detection/diagnosis (CAD/CADx) systems are one of the numerous major research topics in diagnostic radiology and medical imaging. CAD systems allow the radiologist to manipulate mammography to highlight certain features that would otherwise be difficult to visualize. One of the most used techniques is the improvement of contrast, which allows to highlight objects in areas of low intensity. To date, CAD is a more suitable method for primary diagnosis of cancer in computed tomography, X-ray, MRI, or mammogram images. CAD system is an effective intermediate between input images and the radiologist. The output from CAD is not considered as an end result; nevertheless, the result is used as

The CAD approach helps medical doctors to diagnose diseases with a higher degree of efficiency, while minimizing examination time and cost, as well as avoiding unnecessary biopsy procedures. However, CAD systems not only allow a better visualization of mammograms, but also using different digital image processing (DIP), knowledge discovery from data (KDD), artificial intelligence (AI) techniques such artificial neural networks (ANN) allow to

Classification of histopathology images into distinct histopathology patterns, corresponding to the noncancerous or cancerous condition of the analyzed tissue, is often the primordial goal in image analysis systems for cancer automatic-aided diagnosis applications. Recent advances in DIP, KDD, and AI techniques allow to build CAD/CADx that can assist pathologists to be more productive, objective, and consistent in diagnosis. The main challenge of such systems is

The aim of this research is to use advanced DIP to investigate and develop specific imaging biomarkers for Mexican patients through the quantitative mammography analysis and with this information to develop technology based on advanced KDD and AI techniques, aiming to detect breast cancer in the early stages in order to support the diagnosis and prioritization of

2. Development of a CADx system to identify breast abnormalities in

In this research, the study of BC disease using advanced techniques of DIP, KDD, and AI was carried out in order to develop imaging biomarkers that allow to carry out diverse studies for

Currently, there are no public databases of BC in Latin America or Mexico. Therefore, at first stage, different public mammography databases were used for developing and validating digital image processing algorithms capable to select ROIs from mammograms to extract

digital mammograms images using artificial neural networks

BCD. As is showed in Figure 1, the research was divided in three main stages.

preselect certain regions of interests (ROIs) for later analysis by the radiologist [24].

latest advances in computer vision and their manipulation in digital form [22].

reference with regard to additional testing in the related field [23].

164 Advanced Applications for Artificial Neural Networks

dealing with the inherent complexity of histopathological images.

high-risk patients.

image features used to train a generalized regression artificial neural network (GRANN). The aim was to generate a methodology for the characterization of mammograms and their association with risk factors in BC patients as well as to integrate and to develop the technological tools for mammography analysis for BCD using AI technology. In this work, results obtained at first stage are presented.

Because there are no public databases of BC for Mexican patients, it was proposed to establish two protocols for the acquisition of mammograms. The first protocol, second stage, seeks to obtain data retrospectively, which will allow obtaining the mammograms necessary to validate in Mexican patients the methodology and technological tools developed at stage one. The aim of the second stage is the generation of an anonymous database of Mexican patients for free use by the scientific community for the study of BC in Mexican patients and to validate the methodologies developed in collaboration with the General Hospital of Zacatecas (GHZ) and the Molecular Medicine Laboratory (MML) from Autonomous University of Zacatecas, Mexico.

In the second protocol, third stage, it is sought to generate a long-term prospective protocol, which will allow the creation of a database with different risk factors associated with the development of BC. This protocol will allow to collect clinical data of patients with both high and low probabilities of developing cancer. These data will be able to validate the methodologies of cancer detection by the scientific community.

The generation of a prospective protocol will allow the expansion of the database for the study of breast cancer in Mexican patients. Unlike the retrospective protocol, prospective protocol aims to include clinical data, risk factors, and mammograms, among others. This database would present to the scientific community a reference for the development of new breast cancer detection techniques in Mexican patients.

#### 2.1. Methodology: feature extraction and neural network training

Patient prioritization can play a very important role in the reach of health services in developing countries such as Mexico, where not all have access to these specialized oncology services. Therefore, this research seeks the study of BC by generating a methodology that allows the detection of patients with a high probability of BC. In this work, a new technology to generate mammographic biomarkers and a CADx system for breast cancer diagnosis was designed in order to analyze Digital Image Mammograms (DIM). With this knowledge, it is proposed to create a biomarker specifically designed for the Mexican population. As is showed in Figure 2, the first stage was divided into six main stages.

Figure 2. First stage of research. Mammographic feature extraction and artificial neural network training.
