SYNTHETIC DATA GENERATION AND CLASSIFICATION OF HISTOPATHOLOGICAL IMAGES

Histopathology involves the analysis of microscopic tissue images for diagnosing and studying the progress of diseases, such as cancers. Recently, Artificial Intelligence algorithms reached encouraging success in diagnosing diseases related to these medical images. However, research in this area can be hampered by several problems. Indeed, due to the sensitive nature of medical data, it is challenging to access real datasets, making it impossible to train Deep Learning models. Moreover, real datasets often contain biases or imbalances that hinder the generalization of the results on new unseen data. Variational Autoencoders are a popular class of probabilistic generative models that enable consistent training and a useful latent representation of the original input. However, there are theoretical and practical obstacles that hinder their generative potential. Here, we consider different approaches to address the challenges of synthetic data generation of histopathology images and discuss the potential impact in improving the performance of diagnosis models.

Automated summary

Histopathology involves analyzing microscopic tissue images for diagnosing and studying diseases. Artificial Intelligence algorithms have shown success in diagnosing diseases related to these medical images. However, challenges such as limited access to real datasets and biases in the data hinder research progress in this area.