Video: characterization of breast biopsies
Alberto Velasco Mata, student awarded in the second edition of the Turing Machine awards. Cojali SL Award for the best Master's Thesis, entitled: "Characterization of regions of interest in microscopic images of breast biopsies“, tutored by Eusebio Angulo Sánchez-Herrera and María Gloria Bueno García. Next, you can see in the following video a summary of the work carried out by Alberto
Summary
The application of machine learning techniques for the detection and classification of tumor regions in breast tissue biopsies, including cancerous areas, represents a reduction in costs and facilitates the diagnosis of experts. Existing detectors in this field of application typically perform a general evaluation of the samples provided, focusing their attention only on their tumor regions. However, other regions that are in the cellular environment of the tumor could show signs of cancer. The study of this factor may be useful for experts due to the peculiarities of the biopsy process, since it may be the case that no cellular areas appear in it. A detection that takes into account the characteristics of the tumor context could represent an improvement over current methods. In addition, it is necessary to consider the usual absence of labeled data due to the difficulty of annotating specific areas of tissue and the variety of histological samples.
The study of how the classification of different types of tissues affects detection, considering a reduced volume of data, implies that it will be necessary to propose models whose strong point is the distinction between images with high similarity between them, as is the case of tissues .
This paper presents deep learning models based on the technique known as contrastive learning, with the aim of analyzing and comparing the performance obtained when classifying various types of tissues in breast biopsies. In addition, the amount of data necessary for these models to work properly is analyzed, and techniques are applied to avoid overfitting them.
