Programa del congreso

In recent years, as artificial intelligence models have grown increasingly complex, the demand for large, high-quality datasets has risen to ensure robust performance. One domain that benefits substantially from machine learning (ML), particularly for tasks such as classification and segmentation, is digital pathology. However, the limited availability of data in this field often restricts the effective deployment of ML models. In this study, we address this challenge by leveraging a dataset of breast tissue biopsies for cancer diagnosis. This task requires high-quality, well-annotated images, which are difficult to obtain due to the time-intensive process of microscopic sampling and manual labeling by experts. To mitigate this limitation, we propose a few-shot learning approach using a StyleGAN2-ADA model trained on small datasets with fewer than 500 images per class. We conducted a quantitative analysis comparing the quality of the generated images with that of real ones. Several image quality assessment metrics were applied, revealing no statistically significant differences and thus supporting the suitability of the synthetic images for subsequent ML tasks. Finally, the synthetic images were integrated into the training of a ResNet-50 classifier, achieving results comparable to those obtained with real data alone and even improving sensitivity and precision in specific classes, thereby confirming the effectiveness of GAN-generated data in digital pathology.