Aprendizaje automático para la clasificación de biopsias de próstata
Loading...
Official URL
Full text at PDC
Publication date
2025
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
Este proyecto tiene como objetivo la aplicación de técnicas de aprendizaje profundo, en particular, redes neuronales convolucionales (CNN) para resolver un problema de clasificación de imágenes médicas. En concreto, clasificar un conjunto de imágenes histopatológicas de cáncer de próstata etiquetadas en: “cáncer” y no cáncer”. El trabajo consta de una primera parte donde se estudian y se da una visión teórica de la inteligencia artificial y las redes neuronales. Conceptos claves, principios básicos de redes neuronales artificiales, sus tipos y aplicaciones. Además, un primer caso de uso, una red sencilla para el reconocimiento de dígitos numéricos a través de imágenes con un 98% de precisión en sus resultados. Finalmente, se propone el modelo de estudio para el cáncer de próstata. Un primer modelo, una red capaz de clasificar nuestro conjunto de datos en dos clases: “cáncer” o “no cáncer” en función de si la imagen presenta células cancerosas o no, obteniendo una precisión del 94%. Un segundo modelo, basado en el índice de clasificación de Gleason en cuatro etiquetas y obteniendo un resultado del 80% de precisión en el modelo.
This project aims to apply Deep Learning techniques, particularly convolutional neural networks (CNN), to solve a medical image classification problem. Specifically, the classification of a set of histopathological images of prostate cancer labeled as “cancer” and “no cancer.” The work consists of an initial section that studies and provides a theoretical overview of artificial intelligence and neural networks. Key concepts, basic principles of artificial neural networks, types, and applications are covered. Additionally, a preliminary use case is presented: a simple network for recognizing numeric digits in images, achieving 98% accuracy in its results. Finally, the proposed study model for prostate cancer is presented. The first model is a network capable of classifying the dataset into two classes: “cancer” or “no cancer,” depending on whether the image shows cancerous cells or not, achieving 94% accuracy. A second model is based on the Gleason classification model, categorizing into four labels and achieving 80% accuracy in the results.
This project aims to apply Deep Learning techniques, particularly convolutional neural networks (CNN), to solve a medical image classification problem. Specifically, the classification of a set of histopathological images of prostate cancer labeled as “cancer” and “no cancer.” The work consists of an initial section that studies and provides a theoretical overview of artificial intelligence and neural networks. Key concepts, basic principles of artificial neural networks, types, and applications are covered. Additionally, a preliminary use case is presented: a simple network for recognizing numeric digits in images, achieving 98% accuracy in its results. Finally, the proposed study model for prostate cancer is presented. The first model is a network capable of classifying the dataset into two classes: “cancer” or “no cancer,” depending on whether the image shows cancerous cells or not, achieving 94% accuracy. A second model is based on the Gleason classification model, categorizing into four labels and achieving 80% accuracy in the results.












