One-Shot Tumour Synthesis for PET-CT Imaging

Accurate lesion segmentation in whole-body PET/CT scans is crucial for cancer diagnosis and treatment planning, but limited datasets often hinder the performance of automated segmentation models. The use of deep prior from a generative model for tumour synthesis onto healthy images as a means of data augmentation for segmentation tasks is a recent solution to the problem of limited medical image datasets. However, the application of this approach on PET-CT images has not been explored before. To resolve the challenge, we introduce a novel approach which conditions the diffusion process of our data augmenter (which is a diffusion model) with not only the segmentation mask but also reference tumour images. Our findings show that the model trained on the dataset augmented by our data augmenter achieves a higher Dice score than that on the original dataset, demonstrating the potential of our data augmentation approach. In a nutshell, this work presents a promising direction for improving lesion segmentation in whole-body PET/CT scans with limited datasets, potentially enhancing the accuracy and reliability of cancer diagnostics.