The field of medical imaging and conformal prediction is rapidly advancing, with a focus on improving the accuracy and reliability of image segmentation and uncertainty quantification. Recent developments have led to the creation of new frameworks and methods that can efficiently handle high-dimensional data and provide robust predictions. Notably, the use of physics-informed neural networks and conformal prediction has shown great promise in addressing the challenges of medical imaging.
One of the key directions in this field is the development of methods that can adapt to example difficulty, providing larger prediction sets for more difficult examples and smaller ones for easier examples. This has led to the creation of new metrics and algorithms that can better evaluate adaptiveness and provide more accurate predictions.
Another important area of research is the development of training-free frameworks for robust out-of-distribution tumor segmentation. These frameworks operate via a two-stage reason-and-reject process, employing anatomical reasoning and statistical rejection to localize organ anchors and generate multi-scale ROIs.
Some noteworthy papers in this area include: PINGS-X, which proposes a novel framework for super-resolution of 4D flow MRI data using axes-aligned spatiotemporal Gaussian representations. Quantifying and Improving Adaptivity in Conformal Prediction through Input Transformations, which introduces a new metric to evaluate adaptiveness in conformal prediction. R$^{2}$Seg, which presents a training-free framework for robust OOD tumor segmentation via anatomical reasoning and statistical rejection. Minimax Multi-Target Conformal Prediction with Applications to Imaging Inverse Problems, which proposes an asymptotically minimax approach to multi-target conformal prediction. Hierarchical Semantic Learning for Multi-Class Aorta Segmentation, which addresses the challenges of multi-class aorta segmentation using a curriculum learning strategy and a novel fractal softmax. Improving segmentation of retinal arteries and veins using cardiac signal in doppler holograms, which proposes a simple yet effective approach for artery-vein segmentation in temporal Doppler holograms. Controlling False Positives in Image Segmentation via Conformal Prediction, which introduces a simple post-hoc framework that constructs confidence masks with distribution-free, image-level control of false-positive predictions.
