The field of medical image segmentation is rapidly evolving, with a focus on developing innovative methods to improve accuracy, efficiency, and adaptability. Recent studies have explored the use of progressive test-time energy adaptation, mutual aid between foundation and conventional models, and interactive segmentation with reference guidance. These approaches have shown promising results in addressing challenges such as distribution shifts, overconfident predictions, and interactive ambiguity. Notably, some papers have introduced novel frameworks, such as Synergistic training, RefCut, and ADZUS, which have demonstrated superior performance on various medical image segmentation tasks. Furthermore, researchers have investigated the use of self-attention diffusion models, prior-guided SAM, and lightweight GAN-based approaches to improve segmentation accuracy and reduce computational costs. Noteworthy papers include: RefCut, which introduces a reference-based interactive segmentation framework to address part ambiguity and object ambiguity in segmenting specific targets. ADZUS, which leverages self-attention diffusion models for zero-shot biomedical image segmentation and achieves state-of-the-art performance on various medical imaging datasets. BiPrompt-SAM, which presents a novel dual-modal prompt segmentation framework that fuses the advantages of point and text prompts through an explicit selection mechanism, achieving comparable performance to state-of-the-art specialized medical segmentation models.
Advances in Medical Image Segmentation
Sources
Progressive Test Time Energy Adaptation for Medical Image Segmentation
Steady Progress Beats Stagnation: Mutual Aid of Foundation and Conventional Models in Mixed Domain Semi-Supervised Medical Image Segmentation
RefCut: Interactive Segmentation with Reference Guidance
Self-Attention Diffusion Models for Zero-Shot Biomedical Image Segmentation: Unlocking New Frontiers in Medical Imaging
PG-SAM: Prior-Guided SAM with Medical for Multi-organ Segmentation
LGPS: A Lightweight GAN-Based Approach for Polyp Segmentation in Colonoscopy Images
OCRT: Boosting Foundation Models in the Open World with Object-Concept-Relation Triad
MODIS: Multi-Omics Data Integration for Small and Unpaired Datasets
Out-of-distribution evaluations of channel agnostic masked autoencoders in fluorescence microscopy
Show and Segment: Universal Medical Image Segmentation via In-Context Learning
VGAT: A Cancer Survival Analysis Framework Transitioning from Generative Visual Question Answering to Genomic Reconstruction
Flow to Learn: Flow Matching on Neural Network Parameters
BiPrompt-SAM: Enhancing Image Segmentation via Explicit Selection between Point and Text Prompts
Flip Learning: Weakly Supervised Erase to Segment Nodules in Breast Ultrasound
A weakly-supervised deep learning model for fast localisation and delineation of the skeleton, internal organs, and spinal canal on Whole-Body Diffusion-Weighted MRI (WB-DWI)
Benchmarking and optimizing organism wide single-cell RNA alignment methods
AMA-SAM: Adversarial Multi-Domain Alignment of Segment Anything Model for High-Fidelity Histology Nuclei Segmentation
Test-Time Visual In-Context Tuning
iMedImage Technical Report
MO-CTranS: A unified multi-organ segmentation model learning from multiple heterogeneously labelled datasets