Abstract
Biomedical engineering has been targeted as a potential research candidate for machine learning applications, with the purpose of detecting or diagnosing pathologies. However, acquiring relevant, high-quality, and heterogeneous medical datasets is challenging due to privacy and security issues and the effort required to annotate the data. Generative models have recently gained a growing interest in the computer vision field due to their ability to increase dataset size by generating new high-quality samples from the initial set, which can be used as data augmentation of a training dataset. This study aimed to synthesize artificial lung images from corresponding positional and semantic annotations using two generative adversarial networks and databases of real computed tomography scans: the Pix2Pix approach that generates lung images from the lung segmentation maps; and the conditional generative adversarial network (cCGAN) approach that was implemented with additional semantic labels in the generation process. To evaluate the quality of the generated images, two quantitative measures were used: the domain-specific Frechet Inception Distance and Structural Similarity Index. Additionally, an expert assessment was performed to measure the capability to distinguish between real and generated images. The assessment performed shows the high quality of synthesized images, which was confirmed by the expert evaluation. This work represents an innovative application of GAN approaches for medical application taking into consideration the pathological findings in the CT images and the clinical evaluation to assess the realism of these features in the generated images.

Abstract

Abstract

Abstract
Deep learning in computational pathology (CPath) has rapidly advanced in recent years. Research has primarily focused on enhancing accuracy and interpretability across various histology image analysis tasks, from tile-level to slide-level foundation models and novel multiple instance learning (MIL) strategies. However, it is equally important for models to provide well-calibrated confidence estimates. Due to factors such as dataset bias, overfitting, and limited training data, existing models tend to be overly confident on test sets. Promising solutions to address this issue include temperature scaling, a post-hoc method that adjusts logits using a single scalar value. However, the role of calibration in CPath is yet to be clarified. In this study, we evaluate temperature scaling and linear temperature scaling for CPath tasks, analyzing their impact on recalibration in both in-domain and out-of-domain distributions. The results show the limitations of current probability calibration techniques and motivate future work. © 2025 IEEE.