Lung Segmentation

healthy lung segmentation

Unsupervised 3D lung segmentation. Powered by the unique "minimal model" technology that uses a handful of annotated 2D images to understand the key structural properties of the organ in 3D, this new system computes precise segmentations of previously unseen 3D data-sets in seconds!

Featuring precision, scalability, speed and low cost, this solution is the ideal tool for big data analytics, unlocking the contents of massive image archives and beyond.

COVID-19 opacity in 2D images (CT scans)

Precise lung segmentation in the presence of lung opacities that 'leak' into muscle tissue

Developed a fully automated protocol for the detection and extraction (image segmentation) of the Coronavirus (COVID-19) symptoms from lung CT-scans of infected people.

First image: Symptoms of mid-stage incubation. Credit: Fang Y, Zhang H, Xie J, et al. Sensitivity of Chest CT for COVID-19: Comparison to RT-PCR. Radiology doi: 10.1148/radiol.2020200432. Published online February 19, 2020. © Radiological Society of North America.

Second image: segmented symptoms (ground glass) overlaid with the original image

COVID-19 opacities in 3D datasets (CT scans)

Semi-supervised segmentation of lungs with opacities due to COVID-19 infection. The method works by segmenting the pair of lungs from CT scans using the "minimal model" technology. Training a minimal model requires no more than 100 2D annotated images of various cross-sections of the lungs. The system learns critical structural information that can then be used to drive the automated segmentation in 3D. Lungs with ground glass touching the walls are improperly segmented. The system rectifies missing information using a sophisticated module based on 3D connected operators.

Image 01: a cross section of a pair of the lungs from a patient infected by COVID-19; The bright, highly textured area at the bottom of the left lobe is a trace of lung fibrosis, referred to as ground glass in this case. Credit RAIOSS, Brazil

Image 02: a large cluster of ground glass at the edge of the lung lobe. Default visualization settings;

Image 03: a minor cluster of ground glass at the edge of the lung lobe. Default visualization settings;

Image 04: enhanced rendering of the larger cluster, suppressing minor vascular structures.

Advancing the state of the art in understanding coronavirus. This works address

- automated lung detection (check),

- high quality automated 3D lung segmentation,

- and automated segmentation and quantification of ground glass clusters for monitoring the progression of the symptoms!

Image slideshow:

1. 2D slice of CT scan showing ground glass within the lungs

2. the same slice after the 3D segmentation of the lungs

3. the lung volume set rendered in 3D

4. including a transparency layer

5. visualizing the segmented ground glass clusters within the lungs

Lung Cancer Segmentation from PET-CT data

Supervised segmentation of lung tumors from registered CT and PET data sets. The method uses the minimal model technology for the segmentation of lungs from the CT data. This in turn is used as a spatial constraint in the segmentation of 'anomalies' in the PET data set. Anomalies that are observed within the lung area are classified using a neural network classifier for rejecting false positives. The method delivered >96% precision/recall on 120 test cases.

brain trauma and aneurysms

brain hemorrhage

Brain hemorrhage segmentation

Brain hemorrhage segmentation - challenging examples

angiography

blood vessel clean-up

Blood vessels are often imaged in CT scans with the aid of contrast agents. CT scans may be noisy and the application of regular noise removing operators may distort the blood vessels or even remove smaller disconnected, due to the image acquisition method, vessel segments.

aneurysms

Aneurisms - Wikipedia

An aneurysm is a localized, blood-filled balloon-like bulge in the wall of a blood vessel.[1] Aneurysms can occur in any blood vessel, with particularly lethal examples including aneurysms of the Circle of Willis in the brain, aortic aneurysms affecting the thoracic aorta, and abdominal aortic aneurysms. Aneurysms can arise in the heart itself following a heart attack, including both ventricular and atrial septal aneurysms.

As an aneurysm increases in size, the risk of rupture increases.[2] A ruptured aneurysm can lead to bleeding. Aneurysms are a result of a weakened blood vessel wall, and can be a result of a hereditary condition or an acquired disease. Aneurysms can also be a nidus for clot formation (thrombosis) and embolization. The word is from Greek: ἀνεύρυσμα, aneurysma, "dilation", from ἀνευρύνειν, aneurynein, "to dilate".

Diagnosis of a ruptured cerebral aneurysm is commonly made by finding signs of subarachnoid hemorrhage on a computed tomography (CT) scan.

stenosis

Stenosis - Wikipedia

A stenosis (/stəˈnoʊsɪs/;[1][2] plural: stenoses, /stəˈnoʊˌsiːz/; from Ancient Greek στενός, "narrow") is an abnormal narrowing in a blood vessel or other tubular organ or structure.

The example above demonstrates the detection of blood vessel stenosis in the patient's brain.

the brain

Brain Morphology

There exist many medical conditions that may be diagnosed from the brain morphology. These two examples showcase the segmentation of the hippocampus form MRI (above) and CT (below) scans.

osteopathy

Knee

The first case shows the patient's knee with anterior tibial osteotomy. High tibial osteotomy is an orthopedic surgical procedure which aims to correct a varus deformation with compartmental osteoarthritis.

The data set is a CT scan. The exercise involves segmenting the bone mater with high precision.

arthritis

Arthritis is the swelling and tenderness of one or more of the patient's joints. The main symptoms of arthritis are joint pain and stiffness, which typically worsen with age. The most common types of arthritis are osteoarthritis and rheumatoid arthritis.

skull puncture

This exercise shows a young patient with a skull puncture adjacent to the nasal cavity.

anatomy

Chest

The first case shows an undisclosed condition of the patient's chest. The data set is a CT scan. The exercise involves filtering out the noise to optimize visualization of the exterior of the lungs.

stent - aorta

The second case shows an aortic stent. The data set is a CT scan. The exercise involves segmenting the aorta and stent for unobstructed visualization.

Colon

The third case shows an attempt to visualize the internal walls of the patient's colon. The segmentation in incomplete but shows a rather large proportion of the colon

urology

This section showcase exercises aiming at the segmentation of kidney stones found in the urinary tract and kidneys.

kidney stones

The two images above show the kidney stone segmented and superimposed with with the subject's skeleton for localization. The attribute filters are implemented on the Max-Tree data structure and are sensitive to sphericity, compactness and size.

The images above illustrate the smoothing effect of increasing the iso-surface threshold and teh distortion caused by sparse slice spacing.

Extreme case with major stones obstructing the function of both kidneys. A stent is visible along with substantial aortic calcifications.

urinary tract

The forth case shows a cluster of three stones discovered in the subject's urinary tract.

The patient was scanned for a severe condition on one of the hip joints and three renal calculi were detected traveling down the urinary tract (shown in red above).

Bladder stone

The fifth case shows an unusually large prostate stone.

microscopy

The first case shows a 3D view of a neuron acquired using confocal microscopy techniques. The exercise demonstrates a shape preserving noise removal using connected attribute filters. The objective is to remove all speckle without loosing thin disconnected segments presumed to belong to the neuron.