Divide Then Diagnose
PulseAugur coverage of Divide Then Diagnose — every cluster mentioning Divide Then Diagnose across labs, papers, and developer communities, ranked by signal.
7 day(s) with sentiment data
-
PET segmentation models' overlap metrics fail to certify patient-level tumor burden
A new research paper published on arXiv explores the effectiveness of overlap metrics, such as Dice, in evaluating PET segmentation models for estimating metabolic tumor volume (MTV). The study found that while overlap …
-
New causality-based metric improves histopathology image segmentation reliability
Researchers have introduced a new evaluation framework for histopathology image segmentation models, focusing on causality to ensure models rely on relevant pathological structures rather than spurious correlations. Thi…
-
New benchmark KU-PCPE tests AI's ability to ground visual evidence for image composition
Researchers have introduced KU-PCP Elements (KU-PCPE), a new benchmark designed to evaluate how well models can identify the specific visual regions within an image that justify a given photographic composition label. T…
-
New research highlights speed-safety trade-off in text-to-image models
A new research paper titled "Haste Makes Waste? Towards Safe Few-Step Text-to-Image Models" published on arXiv explores the trade-off between speed and safety in text-to-image generation models. The study reveals that f…
-
Few-step text-to-image models trade speed for safety, new paper finds
A new paper from Hugging Face researchers explores the trade-off between speed and safety in few-step text-to-image models. They found that faster generation times in these distilled models often lead to higher rates of…
-
AI-generated satirical video uses face-swapping technology
A satirical video created using AI technology features a face swap, indicating it is not a genuine recording. The content, shared on Mastodon, is presented as humor and parody.
-
TRIP-Shapley framework improves P2P federated learning attribution
Researchers have developed TRIP-Shapley, a new framework for attributing contributions in peer-to-peer federated learning (P2PFL). This method addresses communication overhead and misreporting attack vulnerabilities inh…
-
New MID metric evaluates medical image segmentation distributions
Researchers have introduced MID (Mask-Image Distributional Divergence), a novel evaluation metric for medical image segmentation that operates at the distribution level rather than sample-wise comparisons. This new metr…
-
Retinal vessel segmentation accuracy affected by threshold selection, study finds
A new research paper evaluates the impact of observer choice and threshold selection on retinal vessel segmentation accuracy. The study utilized the CHASE DB1 dataset and analyzed different thresholding policies, includ…
-
New FCA-Guided framework offers perfect validity for AI breast cancer diagnosis explanations
Researchers have developed a novel framework called FCA-Guided Counterfactual (FCA-CF) to generate actionable explanations for multi-modal breast cancer diagnosis models. This framework uses Formal Concept Analysis to c…
-
New BSC-Net improves coronary vessel segmentation for disease analysis
Researchers have developed BSC-Net, a novel framework based on ResNet-U-Net designed to improve the segmentation of small coronary vessels in X-ray coronary angiography. The network addresses challenges like imaging noi…
-
New XAI method generates perceptible counterfactual examples using expert knowledge
A new research paper introduces DiCEf, an extension of the DiCE method for generating counterfactual examples (CFEs) in explainable AI (XAI). This enhanced method integrates expert knowledge through a fuzzy linguistic v…
-
DICE researcher presents ASTRA paper on knowledge graph embeddings at ECML PKDD
Duygu Ekinci Birol, a researcher at DICE, presented the paper "ASTRA: Adaptive Structure-Aware Post-Hoc Alignment of Knowledge Graph Embeddings" at ECML PKDD 2026 in Naples. The research, co-authored by Mohamed Sherif, …
-
New SpFiLM technique enhances brain MRI parcellation accuracy
Researchers have developed a new technique called Spatial Feature-wise Linear Modulation (SpFiLM) to improve the accuracy of automated brain parcellation, particularly for contrast-enhanced T1ce MRI scans. Traditional m…
-
New methods improve fiber bundle segmentation in brain histology
Researchers have developed new methods for segmenting fiber bundles in tracer histology data, a crucial step for understanding brain connectivity. The study compares traditional pixel-overlap losses like BCE and Dice wi…
-
DICE technique enhances text-to-image generation by refining embeddings
Researchers have developed a new technique called DICE (Distilling Classifier-Free Guidance into Text Embeddings) to improve text-to-image generation. DICE refines text embeddings to mimic the effects of classifier-free…
-
AI model performance and human variability in cancer index assessment analyzed
Researchers have evaluated how variations in human interpretation and AI model performance affect the assessment of radiological Peritoneal Cancer Index (rPCI) scores using contrast-enhanced CT scans. The study found th…
-
New framework automates medical imaging code generation
Researchers have developed AutoMedImg, a novel multi-agent framework designed to fully automate the generation of medical imaging processing code. This system operates in two phases: planning, which includes dataset ana…
-
New AI method improves MRI segmentation reliability across domains
Researchers have developed a new method called CARD (Calibration via Agreement in Reverse Diffusion) to improve the reliability of AI segmentation models in medical imaging, particularly when dealing with out-of-domain …
-
New FCom-DICE method conceals communities from graph neural networks
Researchers have developed a new method called FCom-DICE to protect sensitive communities within networks from being identified by graph neural networks (GNNs). This technique involves making small, utility-preserving m…