Acknowledgements

The Computer Aided Diagnosis system consists of two main stages: the first one detects the suspicious regions with high sensitivity presenting the results to the radiologist aiming to reduce false positives. This process is given in the first instance by a preprocessing algorithm based on advanced DIP techniques designed to reduce the noise acquired in the image and in an improvement of the same, and then it executed a segmentation process of different ROIs designed to detect high suspicion of some signs of cancer. By using the information obtained in the segmentation process, the classification of positives or negatives prediction of BC is obtained through a GRANN. After concluding the development of this CADx technological computer tool, it is planned to be used at real workplaces such as GHZ, making a validation of the prediction obtained by the neural network compared with predictions made by specialized

oncologists aiming that it can be used as an aid in the early breast cancer diagnosis.

of malignancy, time of presence, etc.

176 Advanced Applications for Artificial Neural Networks

3. Conclusions

system for BCD.

The aim of developing this CADx system for Mexican patients is to expand the knowledge of the database of BCD including image mammograms information, clinical data, risk factors, biopsy results, genomic information, etc., as is showed in Figure 9, at the bottom of the main window, with the aim to obtain more information on the resulting diagnostics such as degree

As mentioned before, conventionally, BCD and classification is performed by a clinician or a pathologist by observing stained biopsy images under the microscope. However, this is time consuming and can lead to erroneous results. Therefore, there is a rise in the need for developing intelligent and automated technology. As this problem can be computationally modeled as a problem of retrieving relevant information from a plethora of reports or tests, the development of a computationally efficient and detection-wise effective system for BCD can be seen as an issue from the field of KDD. Being cross-dimensional, KDD uses algorithms and techniques from a vast array of fields like soft computing, pattern recognition, machine learning statistics,

In this research, a method based on advanced DIP, KDD, and AI techniques was used for the extraction of fundamental features in DIM in order to detect breast lesions by using a GRANN. The used methodology was divided in two main stages. The first one used advanced DIP techniques for extracting image features of DMI in order to create a biomarker for BCD. With this information, in the second stage, a GRANN was trained and tested in order to classify BC.

After 2000 network trainings, a smoothing factor equal to 1e4 was calculated. This value was used for training the neural net reaching an accuracy of 95.83%. The performance of trained GRANN was evaluated using four performance measures: accuracy, sensitivity, specificity, and precision. The results obtained in this work show that GRANN is a promising and robust

In a third stage, a CADx system based on AI technology is being designed in order to be applied in Mexican patients in collaboration with GHZ and MML. The proposed system aims to eliminate the unnecessary waiting time as well as reducing human and technical errors in

artificial intelligence (AI), natural language processing (NLP), etc.

This work was partially supported by Fondo Sectorial de Investigación para la Eduación under contract no. 241771 and PRODEP under contract no. 511-6/17-7763. The second and sixth authors want to thank the postdoctorate and doctorate scholarships, with scholarship holder numbers 25467 and 25976, respectively, received by Fondo Sectorial de Investigación para la Eduación under contract no. 241771. The fourth and fifth authors want to thank CONACYT PNPC scholarships, with scholarship holder numbers 285593 and 577554, respectively, received through Posgrado en Ingenieria y Tecnologia Aplicada, member of Programa Nacional de Posgrados de Calidad (PNPC) of CONACYT, México.

Authors want to thank the participation of Arturo Serrano Muñoz, an undergraduate student of Information Technology and Communication Bachelor from Universidad Tecnologica del Estado de Zacatecas, México, who is receiving a scholarship from PRODEP under contract no. 511-6/17-7763. This student performed the installation and configuration of the software and several services on the server where calculations were performed for extracting the features of the mammographic images and the neural network trainings.

Authors want to thank the participation of Edgar Viveros Llamas, an undergraduate student of Robotics and Mechatronics Bachelor from Universidad Autonoma de Zacatecas, Mexico. This student is making a social service in Laboratorio de Innovacion y Desarrollo Tecnologico en Inteligencia Artificial, helping with general activities related to this research.
