Loading...

Follow RSIP Vision on Feedspot

Continue with Google
Continue with Facebook
or

Valid

The paranasal sinuses are air-filled spaces surrounding the nasal cavity. The sinuses include the maxillary (purple in the image below), frontal (red), ethmoidal (blue) and sphenoidal (green) sinuses. Due to being air filled, the sinuses make the skull lighter and help with filtration of inhaled air.

Sinus segmentation

Sinusitis is the inflammation of the lining mucous membrane of the sinus cavities, which can be caused by infection (viruses, bacteria or fungi) or allergy. The condition is very common, affecting one in 7-8 people in the United States and other Western countries. Symptoms include nasal mucopurulent discharge, nasal congestion, facial pain and reduction or loss of sense of smell. Persisting of symptoms for longer than 12 weeks is considered chronic sinusitis. The diagnosis of chronic sinusitis is based on clinical symptoms as well as imaging and or endoscopy.

The proximity of the paranasal sinuses (especially the frontal or sphenoid sinuses) to the brain may cause serious complications. These complications can arise due to bacterial infection of the sinuses spreading to the brain via bone and blood vessels. Once the infection reaches the brain, it may lead to meningitis or to formation of brain abscesses. In addition, proximity to the orbit may result in periorbital or orbital cellulitis and abscesses.

Chronic sinusitis significantly reduces the quality of life. In severe cases of sinusitis, surgery is indicated. Surgical alternatives include Functional Endoscopic Sinus Surgery (FESS) and Sinus ostial dilation (balloon ostial dilation). The latter is a procedure in which a balloon catheter is used to dilate the sinus ostium.

Automated sinus segmentation.

RSIP Vision has created a computer vision algorithm to segment the sinuses on computerized tomography (CT). Using our algorithm, the sinus ostium can be identified and utilized for guidance of balloon inserted as part of a sinus ostial dilation. Our algorithm easily separates air, soft tissues and bony structure even under the presence of unique pathologies. Classical algorithms, based on the modeling of the sinuses, are combined with cutting edge deep learning methods into a full module designed to producing the most accurate automated sinus segmentation. Want to know more? Talk now with one of our medical imaging experts.

The post Sinus segmentation with deep learning appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Intracranial hemorrhage is defined as bleeding within the cranium. Bleeding can occur within the brain parenchyma itself, giving way to what is called intra-axial bleed or intracerebral hemorrhage. On the other hand, bleeding outside the brain tissue is referred to as extra-axial bleed. Subtypes of extra axial bleed include epidural, subdural and subarachnoid hemorrhages.

Brain hemorrhage segmentation (in red)

The intracranial hemorrhage might cause a Phenomenon called “mass effect”, meaning that the hemorrhage causes areas of brain tissue or brain structures surrounding it to be compressed and injured, due to elevated pressure in the restricted space within the skull. This could result, together with the edema around it, in elevated intracranial pressure (ICP). Elevated ICP is correlated with poor outcome, and may result in intubation, mechanical ventilation, multiple complications; if left untreated, it can be fatal. Thus, early diagnosis, monitoring and treatment are essential.

Two types of intracranial bleeding that cause hemorrhagic stroke are intracerebral hemorrhage and subarachnoid hemorrhage. Intracerebral or subarachnoid bleed can be accompanied by intraventricular hemorrhage (IVH), which is a bleeding into the brain’s ventricular system. IVH might occur after head trauma or as a result of hemorrhagic stroke. In this case as well, the volume of the bleeding is a predictive factor of prognosis and a valid method of evaluation is needed.

Automated brain hemorrhage segmentation

RSIP Vision’s image analysis algorithm module based on deep learning can rapidly estimate the hemorrhage volume and measure the edematous area around it for both ICH and IVH. Some cases of brain hemorrhage may require ICP monitoring and drainage via intra-ventricular catheter. Our image processing algorithm produces a 3D model of the ventricular system, which can ultimately be useful in guidance of the neurosurgeon during ventriculostomy.

SAH (subarachnoid hemorrhage) segmentation is challenging due to very variable and detailed shape, together with low contrast on CT. RSIP Vision used a large quantity of annotated data to feed a deep learning neural network able to segment all kinds of hemorrhages and accurately measure their volume. Cutting-edge Convolutional Neural Networks (CNN) architecture has been instrumental to achieve accurate results. Call our engineers to make sure that this module is what you are looking for.

The post Brain Hemorrhage Segmentation with Deep Learning appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Dental problems affect people of all ages and ethnic groups, and are common worldwide. With many patients suffering from tooth decay, orthodontic issues and even oral cancer, the need of oral surgeries and other treatments increases constantly. Recent advancements in technology allow safer and more accurate procedures, including better preoperative planning aided by 3D segmentation of the teeth and other dental structures.

3D segmentation of intraoral scanners (IOS) computed tomography (CT) dental scans, or cone beam computed tomography (CBCT), which became a useful replacement for regular CT in dental imaging, could be used when planning dental surgeries, such as prosthodontics surgeries including dental implants, root canals treatments, orthodontic treatments, maxillofacial surgeries and even dental biometrics. For instance, automated segmentation could be handy when planning dental implant surgery by allowing the surgeons to better review and inspect the maxilla and its bone quality, which is essential for implant stability. Individual tooth segmentation helps with orthodontic treatment planning and teeth alignment.

Although it is a promising development, automated dental segmentation may turn out to be quite difficult. For example, segmentation of each individual tooth is tricky due to complex teeth arrangements, variation in shape and similarity to adjacent structures such as the tooth socket. The maxilla is also a difficult structure to segment, because of its complex thin bony components including the palate, maxillary sinuses and parts of the orbit walls. The challenge to the segmentation algorithm consists in finding the exact classification in abnormal teeth structures like overlapping teeth and abnormal growth directions.

Automated Dental Segmentation

RSIP Vision’s engineers developed a module for automatic segmentation of the dental structure. It is based on deep learning neural networks and advanced mathematical algorithms from graph theory.

Using those computer vision and artificial intelligence methods, we created a fully automatic and accurate anatomical model of teeth, gums and jaws. Dental professionals use to get an accurate dental and maxillary map in view of specific diagnosis and treatment.

The post Dental Segmentation with Deep Learning appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Lungs vasculature has a major part in blood oxygenation. The complicated branches of arteries and veins, accompanied by the intricate bronchial tree are in charge of gas exchange, where inhaled air, containing oxygen, enters the lungs and through the bronchial tree it reaches the alveoli. There, oxygen enters the thin capillaries covering the alveolar sac. In a similar manner, carbon dioxide exits the capillaries into the alveoli and exhaled. From the lung, oxygenated blood reaches the left atrium of the heart via the pulmonary veins. Then, the blood enters the left ventricle and into the aorta, which distributes the oxygenated blood to the rest of the body. Deoxygenated blood is carried from the body to the heart by the vena cava, and travels into the lungs through the pulmonary arteries, where it is later oxygenated as mentioned above.

Automated lung vasculature segmentation

Mapping pulmonary blood vessels, major and small, is essential in clinical setting. For example, knowing the vasculature helps with planning of surgeries within the chest cavity and the lungs, including segmentectomy, lobectomy, pneumonectomy or lung biopsy. In addition, knowledge of the vessels branching may help with distinguishing vessels from nodules, masses or other pathologies involving focal opacities. Furthermore, signs of pulmonary hypertension or  pulmonary embolism can be identified by proper segmentation of the vessels. In the case of an embolism, determining which vessel is occluded could help choosing the best treatment possible. Manual tracing of the pulmonary vessels can be extremely time consuming for the radiologist. Even the experienced physician may come across difficulties when trying to trace the smallest vessels, due to image noise. Another challenge is given by the difficulty in distinguishing between lung vessels and  the tumors which are crossed by them: they have different shape but similar intensity and thus are difficult to segment with classical computer vision techniques. RSIP Vision has developed the most accurate AI module for lung vessel segmentation to solve ideally these challenges. Our automated module uses deep learning neural networks to apply the full power of AI algorithms to lung vasculature segmentation. Contact us and we shall discuss with you the fittest solution for your project.

The post Lung vasculature segmentation appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Lung cancer is the most common cancer related mortality cause among men, and second in women worldwide. Primary lung cancer is usually divided into two groups: Non-Small Cell Lung Cancer (NSCLC), and Small Cell Lung Cancer (SCLC), the former being the most common. In cases that are diagnosed late, prognosis might be grim with high mortality rates. However, early diagnosis and treatment can improve outcome. The lung is also a common site for metastatic disease. Common primary tumors which spread to the lungs, include colorectal cancer, breast cancer, renal cancer, melanoma and head and neck cancers.

Lung tumors might appear in imaging as nodules (<3cm), or as masses (>3 cm). In order to identify the type of lung tumor, a biopsy is usually performed. Tissue sample can be obtained transbronchial via bronchoscopy, percutaneous guided by CT, thoracoscopically or via open biopsy. Usually after pathological confirmation of malignancy, the disease extent is classified according to the TNM classification of malignant tumors, and helps determine treatment protocol. A measure called Response Evaluation Criteria In Solid Tumors (RECIST) is used to evaluate treatment response.

Automated lung tumor segmentation

Lung tumor segmentation can assist in diagnosis, staging and follow-up, and can also be used by the physician during biopsy trajectory planning, surgery and image-guided ablation. Segmentation of tumors can be quite challenging, since their shape and size varies a lot. This means, that it is important to gather a large quantity of data to train the deep learning system and also that these data must be accurately annotated by a radiologist. In addition, distinguishing between tumor tissue and other lung findings such as pleural effusion, lymph nodes and consolidation can be difficult due to similar density. In order to deal with this challenge, RSIP Vision has built a proprietary neural network architecture that takes into account both global and local context of the tumor. This cutting edge AI module is today’s best solution for lung tumors segmentation. Contact RSIP Vision’s engineers to learn how.

The post Lung Tumor Segmentation appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

RSIP Vision is introducing Deep Learning in the biopsy analysis procedure. Here is how the benefits of tissue analysis with AI have become available. Biopsy is the key examination used to determine the presence of most malignant tumors. Notwithstanding its importance and the need for millions of examinations every year, the analysis of biopsy in the search of cancer cells has been based for many years on tedious work of expert cytologists. This has exacerbated the demand on this resource, which is always expensive and often scarcely available.

By introducing the revolutionary AI Deep Learning technologies, it has become possible to carry the important and critical tissue analysis with AI in a much more reliable and less human-intensive way.

There are many challenges to applying Deep Learning to biopsy procedures, starting from the detection of cells and the ability to segment clumps of packed cells into individual cells. Following these steps, more issues need to be solved in the classification of the cells.

Classification is highly dependent on the staining procedure, which is very variable from one lab to another and from one probe to the other. This high inter and intra variability makes it quite difficult to obtain objective and univocal results from biopsy analysis.

Automated Tissue Analysis with AI

RSIP Vision provides a solution to all these challenges, developing Artificial Intelligence algorithms for histopathology needs.

The immense breakthroughs given to us by deep learning technology enable us to achieve accuracy levels that were not previously possible, using only the classical methods of traditional computer vision.

The training stage is quite flexible and it can adjust to images from different sources to compensate for the specific tasks at hand.

With our automated application, the future of pharma and the medical imaging fields looks very promising. Diagnosis is becoming more precise, more robust and much faster. In addition, the ability to match specific treatments to each individual patient is becoming more and more realistic.

Do you want to benefit of the Deep Learning revolution? Ask our engineers to discuss with you the implementation of AI within your organization.

The post Tissue Analysis with AI appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

RSIP Vision’s automated detection and tracking of tumors answers one of the most frequent needs in the pharma industry – assessment of patient response to a new drug. This very strong tool saves precious time in the clinical stage when the automated detection and tracking can supply the most accurate measurement in an efficient way. The protocol might require some manual review of the radiologist – though this intervention will be much faster than a fully manual expert annotation.

The immediate advantage of RSIP Vision’s tool is that it presents each target tumor with the right measurement along the treatment period. In this method, the radiologist is only expected to monitor and approve the AI software’s finding.

Since this is not diagnostic for a specific patient and the purpose of the study is to evaluate the response to the tested drug – there is no risk of false negatives. Even in the hypothesis that the system skips one of the tumors in one scan – it still will be able to fix it when detection is done in the next CT – with no harm to any patient.

The AI software is taught on a training database to distinguish between benign nodules and malignant tumors following shape and color characteristics, just like a human annotator would do. Once trained, the software’s classifier distinguishes between nodules and tumors with state-of-the-art accuracy and without any human intervention.

RSIP Vision’s methodology is built on world-class Deep Learning algorithms, which are recognized by our industry and by academic researchers as the best AI technique to obtain optimum accuracy as fast as modern computers allow it, with no bottleneck due to human expert shortage. No external intervention is needed.

Automated detection and tracking of tumors

RSIP Vision’s detection and tracking of tumors is based on convolutional neural networks, the advantages of which are numerous: speed, full automation, no inter and intra difference on all the dataset: the simplest and most effective answer to all kind of problems in medical segmentation. Its robustness and efficiency have already been successfully tested by RSIP Vision’s clients.

Contact us now to discuss how our Deep Learning Artificial Intelligence solutions can help optimize your clinical trials.

The post Detection and Tracking of Tumors appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

RSIP Vision has develop an efficient tool for clinical trials in the pharma industry – the automated RECIST measurement. The gold standard in measuring the evolution of solid tumors, along the treatment with a tested drug, is the RECIST score (Response Evaluation Criteria in Solid Tumors), which is determined by the radiologist after inspecting patient CT slices: the radiologist measures the longest diameter of 2-5 of the largest tumors through the CT slices. In this way, it is possible to track over time the progression of a tumor. The evolution of the RECIST score will show whether the disease has progressed, stabilized or worsened. This estimation process is time-expensive, it requires a human expert, it is mostly visual and as such it is prone to intra and inter errors, especially when tumor shapes are complicated. Radiologist must visually go through several slices before assessing on which slice(s) tumors are larger.

Automated RECIST measurement with AI

RSIP Vision’s algorithm makes all this process more precise and stable. The algorithm is designed to search for the largest diameter in the slices and it chooses on which slices the most significant information is found. In addition, the automated RECIST score can be measured on all lesions in the CT scans, breaking the limit of 2-5 apparently larger lesions. Also, it is possible to repeat the CT scan after time and to register the new results with the significant slices from the previous control. The AI software verifies the status of the tumor in the current scan and compare it to previous scans, offering an objective assessment of the course of the disease. In turn, this result enables to properly evaluate the objective response to the cancer treatment(s) being clinically tested.

The scientific basis of this new automated procedure for RECIST measurement with Deep Learning has been confirmed by a recent academic study published by Tang et al.: Semi-Automatic RECIST Labeling on CT Scans with Cascaded Convolutional Neural Networks. Their study confirms that “system performs more stably and with less variability, suggesting that RECIST annotations can be reliably obtained with reduced labor and time”.

Automated RECIST measurement with Deep Learning, as performed by RSIP Vision and other researchers, is confirmed to produce scores annotations with less variability with shorter use of expert time. This fully automated procedure proves to be very precious along clinical trials aimed at verifying whether a new drug leads to complete response, partial response, stable disease or progressive disease. Contact now RSIP Vision’s engineers to discuss your own project!

The post Automated RECIST Measurement appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Dendritic cells are a type of antigen-presenting cells and have an integral part in the normal functioning immune system, in that they help to initiate primary immune response. Dendritic cells are typically present in tissues that come in contact with the external environment. That includes the skin, the nasal cavity lining, the lungs and parts of the digestive tract.

Segmentation of Dendritic Cells with Deep Learning

RSIP vision has developed a set of artificial intelligence tools to find, detect and track dendritic cells. Due to dendritic cells being an integral part of primary immune response, one might assume that they help the immune system fight against tumor cells thanks to their ability to present tumor antigens. However, tumor-associated dendritic cells are generally badly defective and they even contribute to immunosuppression in malignant diseases. Therefore, investigation and analysis of tumor-associated dendritic cells have become a major field of interest in search for cancer treatments. It is now known that they may have an impact on disease progression and their infiltration to the tumor increases immune response.

RSIP Vision has developed (in cooperation with Tufts Medical Center) computer vision-based tools which have shed light on the behavior of these very important cells in relation to Dry Eye Disease (DED). An AI module was developed for detection of cells using in vivo confocal microscopy (IVCM) of the human cornea. The application was carried out on corneal dendritic cells to assess the inflammation level, by visualizing the density of immune system cells that promote inflammation. Usually, classification of these cells in IVCM images is done manually, which makes it an extremely time-consuming operation which is also prone to interpretive errors based on the experience level of the examiner. Hence, RSIP Vision has developed automated convolutional neural networks (CNN) for detection and quantification of dendritic cells. These CNNs proved to be very efficient in detecting the cells, as they did it with high accuracy, objectiveness and fast. This process eliminates the risk of interpretation errors, the consequences of which could adversely affect timely diagnosis and effective prognosis.

These algorithmic tools are supplied today by RSIP Vision. Contact our engineers now so that your organization can also benefit from these technology breakthroughs.

The post Detection and Segmentation of Dendritic Cells appeared first on RSIP Vision.

Read Full Article
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Pulmonary nodules (AKA lung nodules) are small masses (up to 30mm) of tissue surrounded by pulmonary parenchyma. They are quite common finding on computerized tomography (CT) scans, and although most lung nodules are benign, some are cancerous. Some of the characteristics of the nodules may indicate high suspicion of malignancy such as large size, irregular/spiculated borders or inhomogeneous density. In that case, a tissue biopsy (percutaneous or via bronchoscopy) might be indicated.

Nodules volume and borders assessment are important in cancer diagnosis and staging. CT offers very reliable imaging for evaluation and follow-up of nodule size, growth, and location. It also allows visualization of nodule attenuation (density) and borders.

Nodule segmentation can also assist the physician during the biopsy, whether it is done via bronchoscopy or percutaneously, by helping the physician choose the best trajectory, hence minimizing complications.

Automated lung nodules segmentation

RSIP Vision has developed a state-of-the-art module for detection of lung nodules. Traditional methods based on classic grey level based techniques would not work in this challenging task. The AI methods were used to address various critical issues to distinguish between nodules and blood vessels that have very similar appearance in grey scale. Hence, the appearance of nodules might be very different between the solid ones and the Ground-Glass opacity ones.

The Deep Learning methods are very versatile and include powerful methods like U-Net and Mask-R CNN that can derive maximum information from the scan in the 3D and the environment of the nodule as well. Much care is taken to furnish the most precise annotated images to the system. During the training phase, the weights of the network get adjusted and refined for the specific task at hand. The result is a very accurate lung nodules segmentation with Deep Learning, that can give you much better results than the ones you currently have. Contact now our engineers to learn more about the automated pulmonary nodules segmentation!

The post Lung Nodules Segmentation appeared first on RSIP Vision.

Read Full Article

Read for later

Articles marked as Favorite are saved for later viewing.
close
  • Show original
  • .
  • Share
  • .
  • Favorite
  • .
  • Email
  • .
  • Add Tags 

Separate tags by commas
To access this feature, please upgrade your account.
Start your free month
Free Preview