Denoising Probabilistic Diffusion Models for Synthetic Healthcare Image Generation

Healthcare data are an essential resource in Machine Learning (ML) and Artificial Intelligence (AI) to improve clinical practice, empower patients and enhance drug development with the aim to discover new medical knowledge. In particular, the biomedical imaging analysis plays a important role in the health-care context producing a huge amount of data that can be used to study complex diseases and their evolution in a deeper way or to predict their onsets. In this work we consider an approach based on Denoising Diffusion Probabilistic Models (DDPM) which is a type of generative model that uses a parameterized Markov chain and variational inference to generate synthetic samples that match real data. In particular, we execute a study by training on Malaria images and generating high-quality synthetic samples in order (i) to test the performance of the DDPMs, (ii) to estimate the association between original and synthetic data and (iii) to understand how the natural and human-made environmental factors impact Malaria disease. Finally, we use a well-defined convolutional neural network for classification tasks to assess the DDPM's goodness in generating the synthetic images.