Artificial intelligence model for the prediction of malignant tumors using a set of medical images from mammography studies

Abstract

Currently, the diagnosis of tumors and malignant cells through imaging studies is a great challenge for expert medical personnel, due to the complexity of achieving an early prediction of cancer cells, which would allow to accelerate early medical treatments. Today, technologies have become a fundamental ally for the health sector, specifically the area of artificial intelligence, which has permeated many disciplines, generating important advances. Advances in parallel computing, GPU technology, and deep learning have made real-time image processing easier. The main objective of this research was to generate a deep learning model for the prediction of malignant cells in medical images of diagnosed mammograms. Using the previously trained model based on Faster R-CNN, with the ResNet function extractor. This model works in the Python programming language, using the Tensorflow framework and the OpenCv library. The algorithms were previously trained through the DDSM and MIAS open medical image databases, published on the web. This model not only focuses on recognizing and classifying malignant cells in the image, but also on the location of objects within it, appropriately drawing a bounding box. One of the latent challenges of these models since their inception has been the consumption of computing, but today they have been optimized so much that they allow freezing the pre-trained models by loading them in the memory of the devices, managing to use them in computers without GPUs. As a result, it was found that the Faster R-CNN method with the Resnet 101 extractor offers great advantages of precision and speed when it comes to detecting malignant tumors, studies that can serve as a great contribution to the bets of this algorithm in the health sector.