RESUMO
RESUMEN Fundamento: la detección y clasificación precisa del cáncer de mama mediante el diagnóstico histopatológico es de vital importancia para el tratamiento efectivo de la enfermedad. Entre los tipos de cáncer de mama, el carcinoma ductal invasivo es el más frecuente. El análisis visual de las muestras de tejido en el microscopio es un proceso manual que consume tiempo y depende del observador. Sin embargo, en muchos países, incluido Cuba, es escaso el uso de herramientas software para asistir el diagnóstico. Objetivo: desarrollar una herramienta software para detectar tejido de cáncer de mama, del subtipo carcinoma ductal invasivo, en imágenes histopatológicas. Métodos: la herramienta se implementó en Python e incluye métodos de detección de carcinoma ductal invasivo en imágenes histopatológicas, basados en algoritmos de extracción de características de color y textura en combinación con un clasificador de bosques aleatorios. Resultados: la herramienta de código abierto brinda una serie de facilidades para la lectura, escritura y visualización de imágenes histopatológicas, delineación automática y manual de zonas cancerígenas, gestión de los datos diagnósticos del paciente y evaluación colaborativa a distancia. Fue evaluada en una base de datos con 162 imágenes de pacientes diagnosticados con carcinoma ductal invasivo y se obtuvo una exactitud balanceada de 84 % y factor F1 de 75 %. Conclusiones: la herramienta permitió un análisis interactivo, rápido, reproducible y colaborativo mediante una interfaz gráfica sencilla e intuitiva. En versiones futuras se prevé incluir nuevos métodos de aprendizaje automático incremental para el análisis de imágenes histopatológicas digitales.
ABSTRACT Background: the accurate detection and classification of breast cancer through histopathological diagnosis is of vital importance for the effective treatment of the disease. Among the types of breast cancer, invasive ductal carcinoma (IDC) is the most common. Visual analysis of tissue samples under the microscope is a manual, time-consuming and observer-dependent process. However, in many countries, including Cuba, the use of software tools to assist diagnosis is scarce. Objective: to develop a software tool to detect IDC subtype breast cancer tissue in histopathological images. Methods: the tool is implemented in Python and includes IDC detection methods in histopathological images, based on algorithms for extraction of color and texture features in combination with a random forest classifier. Results: the open source tool provides a series of facilities for the reading, writing and visualization of histopathological images, automatic and manual delineation of cancer areas, management of patient diagnostic data and collaborative remote evaluation. It was evaluated in a database with 162 images of patients diagnosed with IDC, obtaining a balanced accuracy of 84 % and a F1 factor of 75 %. Conclusions: the tool allowed an interactive, fast, reproducible, precise and collaborative analysis through a simple and intuitive graphical interface. Future versions are expected to include new incremental machine learning methods for the analysis of digital histopathology images.
RESUMO
Deep learning is a class of machine learning methods that has been successful in computer vision. Unlike traditional machine learning methods that require hand-engineered feature extraction from input images, deep learning methods learn the image features by which to classify data. Convolutional neural networks (CNNs), the core of deep learning methods for imaging, are multilayered artificial neural networks with weighted connections between neurons that are iteratively adjusted through repeated exposure to training data. These networks have numerous applications in radiology, particularly in image classification, object detection, semantic segmentation, and instance segmentation. The authors provide an update on a recent primer on deep learning for radiologists, and they review terminology, data requirements, and recent trends in the design of CNNs; illustrate building blocks and architectures adapted to computer vision tasks, including generative architectures; and discuss training and validation, performance metrics, visualization, and future directions. Familiarity with the key concepts described will help radiologists understand advances of deep learning in medical imaging and facilitate clinical adoption of these techniques. Online supplemental material is available for this article. ©RSNA, 2021.
