Improved HER2 Tumor Segmentation with Subtype Balancing using Deep Generative Networks
This work addresses the issue of reliable automatic HER2 analysis for medical diagnosis by reducing performance variance between subtypes, though it is incremental as it applies existing generative methods to a specific domain.
The paper tackled the problem of tumor segmentation in HER2-stained histopathology images, which is complicated by class imbalance and histological subtypes, by using deep generative networks to create synthetic images for balancing subtypes. The result was an increase in tumor Dice score from 0.833 to 0.854 and a reduction in variance between subtype recalls.
Tumor segmentation in histopathology images is often complicated by its composition of different histological subtypes and class imbalance. Oversampling subtypes with low prevalence features is not a satisfactory solution since it eventually leads to overfitting. We propose to create synthetic images with semantically-conditioned deep generative networks and to combine subtype-balanced synthetic images with the original dataset to achieve better segmentation performance. We show the suitability of Generative Adversarial Networks (GANs) and especially diffusion models to create realistic images based on subtype-conditioning for the use case of HER2-stained histopathology. Additionally, we show the capability of diffusion models to conditionally inpaint HER2 tumor areas with modified subtypes. Combining the original dataset with the same amount of diffusion-generated images increased the tumor Dice score from 0.833 to 0.854 and almost halved the variance between the HER2 subtype recalls. These results create the basis for more reliable automatic HER2 analysis with lower performance variance between individual HER2 subtypes.