Desarrollo de una librería para la clasificación de imágenes de mamografías en formato DICOM.

Abstract

The breast cancer is one of the main causes of mortality in women, due to lack of informa tion, awareness and above all the omission of controls. Currently, diagnoses are made by trained medical staff; however, there may be the possibility of human errors in the mamographic analy sis. The mothodology used follows the steps such as reading DICOM mammographic images evaluated with Fuzzy C-Means, K-Means, Random Forest, k-Nearest Neighbors, Support Vec tor Machines and Convolutional Neural Networks. Rethinking Machine Learning algorithms written in C/C++ in such a way that they could be integrated inside the library. A dataset of 150 DICOM mammographic images was used, and the resulting data were compared using Matlab to verify the error rate for each algorithm. For the library evaluation the results obtained are, the average error between the library and MatLab was 0.7%, the benchmarking for each technique in CPU was 11.47%, for RAM memory its average use is 300 Mb and the time of use was 101 seconds on average. It is concluded that although the performance of the library is acceptable, it can be optimized by applying pre-processing, segmentation, extraction and feature selection techniques, reducing the computational load.