Image segmentation is the task of labeling an image at the pixel level, and it underpins much of modern medical imaging, autonomous perception, and scene understanding. Here we break down segmentation architectures, loss functions, and evaluation practices, with frequent attention to the medical settings where boundary accuracy and robustness matter most. Each piece traces the method back to its source research.
A radiologist opens a whole spine MRI of a patient with suspected metastatic disease. There are twenty five vertebral bodies and two dozen discs to outline, from the sacrum up to the first cervical vertebra, plus the spinal cord. A…
TEAMS Mamba Snake Traces Organ Contours With Text Prompts Read More »
A dermatologist looking at a high frequency ultrasound scan of skin can read the layers the way a geologist reads rock strata. There is the bright epidermis at the surface, a dark band beneath it that clinicians call the subepidermal…
Physics Guided Synthetic Ultrasound for Skin Segmentation Read More »
Medical image segmentation is the task of coloring in a scan, drawing a precise outline around every organ or lesion so that a clinician can measure it, track it over time, or plan around it. Done by hand it is…
G2RA-Net: Graph Slices Sharpen Medical Image Segmentation Read More »
Ejection fraction, usually shortened to EF, measures the share of blood that leaves the left ventricle during a contraction. A healthy heart ejects somewhere north of half its filled volume with each beat. When that fraction drops, it is often…
The Segmentation Ceiling: Why Heart Masks Don’t Improve Ejection Fraction Prediction Read More »
Some hearts have one working pump where the textbook says there should be two. Children born with single ventricle physiology grow up with a heart that mixes oxygen rich and oxygen poor blood, and a chamber that was supposed to…
SV-Cine Segments Single Ventricle Hearts From Cine MRI Read More »
A segmentation model can score a near perfect Dice number and still be wrong in the one place a hematologist cares about. Draw the outline of a white blood cell a few pixels too wide and the overlap metric barely…
RABR-Net Refines Cell Boundaries That Dice Scores Miss Read More »
Unsupervised image segmentation asks a model to draw meaningful boundaries around objects in a photo without ever being told what those objects are. No labeled training data, no annotated masks, just pixels in and a partition of the image out.…
DeepCut++’s Single Best Fusion Weight Does Not Actually Win Every Task Read More »
Ask a radiology resident to find the spleen on a single unfamiliar scan after seeing only one labeled example beforehand and you are asking for exactly the problem this paper takes on, except the resident here is a neural network.…
Prototype Guided Graph Reasoning for Few Shot Medical Segmentation Read More »
Researchers at the University of Glasgow and the University of Manchester built a shape-guided Bayesian recurrent framework that estimates 3D cardiac motion from cine CMR images without relying on intensity-based image registration — and it outperforms every major method on…
Researchers from CUHK and Harvard Medical School built a training paradigm where the in-training specialist autonomously crafts prompts for a frozen SAM generalist, which returns noisy-but-useful labels across three orthogonal views — while active learning selects the most informative scans…
GM-ABS: SAM-Driven Active Barely Supervised 3D Medical Image Segmentation Read More »