We are exhibiting at SIIM2018! Visit our booth 108 or schedule a meeting at info@oxipit.ai

Better clinical outcomes through artificial intelligence-powered medical image analysis


ChestGlass [explainer video] is a search engine that finds similar-looking radiographs in a given database.

The similarity is identified by a neural network, which judges both the pathology present as well as other features in the image (such as the location of the pathology, its severity etc.).

When implemented into a PACS/local system/cloud service ChestGlass empowers the user to quickly find retrospective cases with similar radiological appearance. This allows to study the report/final diagnosis/outcome of the patient.


  • workflow: facilitate the diagnostic process and improve accuracy
  • teaching: reduce the learning curve for radiology residents
  • research: find cases that stand out

ChestEye [explainer video] is a fully automatic computer-aided diagnosis (CAD) solution. It currently supports 49 radiological features, and is able to localize these features on a radiogram in terms of a heatmap. ChestEye is under constant improvement with the goal to generate preliminary reports, that incorporate all the radiologically relevant information present in a chest X-ray radiograph.

The following pathologies are supported: atelectasis, enlarged heart, edema, pleural effusion, pulmonary emphysema, subcutaneous emphysema, consolidation, pneumothorax, hypoventilation, lymphadenopathy, hypertension, calcification, elevated diaphragm, dislocated mediastinum, widened mediastinum, congestion, fibrosis, pleural adhesion, abnormal hilar, enlarged hilar, mass, hernia, lung removal, enlarged aorta, goitre, aortic sclerosis, respiratory distress syndrome, pleural thickening, pneumomediastinum, pericardial effusion, pleural plaque, pneumoperitoneum, bone fracture, osteochondrosis, spondylosis, osteoporosis, spinal deformities, and radiological signs of tuberculosis and sarcoidosis.

The following other features and artifacts are supported: central venous catheter, intubation, chest tube, sternal wires, stent graft, artificial heart valve, intra aortic balloon, ventricular assist device, pacemaker.

Integration with several PACS/DICOM viewers is under way. Cloud and local (non-cloud) deployment is also possible.

ChestEye is applicable to chest X-ray PA and AP images. LAT image support is in development.

Performance: area under curve (AUC) averaged over the 49 features is 0.91.

QEye is a radiological queue management/patient prioritization solution, which automatically prioritizes potentially unhealthy patients. Currently, queues of images awaiting description by the radiologist can be sorted by exam time, patient name, modality, the imaging site (device), and possibly a priority column, where the priority is currently set by a radiographer. In our proposed solution, the priority column would be provided by our QEye solution (see figure, indicated by a red rectangle).

Oxipit solution for automatic analysis of brain lesions in head MRI scans: automatic segmentation and volumetric evaluation

  • reduces the time necessary for a radiologist to spend on each case
  • reduces interpersonal variability
  • aids in progression tracking
  • provides the possibility for a parameter-based case search

Oxipit brain ischemic area segmentation solution analyses native head CT scans: automatic segmentation and volumetric evaluation of ischemic areas

The algorithm operates in two stages. First, it segments gray matter structures in native head CT scans. It then uses this gray matter map to detect and segment the ischemic areas.

Problem and Focus

Medical imaging (radiology) is a highly-trained-labour intensive profession. This results in

Our deep learning (also known as artificial intelligence, machine learning, artificial neural networks) expertise and data from our partners (hospitals and healthcare companies) allow us to alleviate these problems by reducing routine work and increasing diagnostic accuracy.

Our Approach

Most medical imaging startups attempt to productize a single medical task. By contrast, we are able to build multiple accurate models with speed and efficiency, thus addressing many different medical problems. This is possible due to our proprietary deep learning platforms and a stellar team.


Our cross-disciplinary team boasts several gold medals in various data science competitions, in addition to several decades of combined computer vision product development expertise. See below and the media section for more info.


Gediminas Pekšys , CEO

  • BA of Mathematics, University of Cambridge
  • 5 years of experience in data science, computer vision and Deep Learning
  • 1st place in Deep Learning competition in Kaggle
  • Deep Learning paper review organizer

Jonas Bialopetravičius

  • MSc in Computer Science, Vilnius University
  • PhD student in Astrophysics, focusing on Deep Learning applications
  • 7 years of experience in computer vision and machine learning in the field of biometrics
  • Two 1st places in Deep Learning competitions in Kaggle

Darius Barušauskas

  • MSc in Econometrics, Vilnius University
  • 7 years of experience in machine learning and Deep Learning applications
  • Created over 30 models for financial, utilities & telco companies
  • International acknowledgement in online data science community
  • 5th overall rank in Kaggle, 5 prize finishes including three 1st places
  • Kaggle Grandmaster

Naglis Ramanauskas

  • MSc in Medicine, Vilnius University
  • Resident doctor of Radiology, Vilnius University
  • Deep Learning experience in several radiology tasks
  • 1st place in 2017 AI Hackathon Vilnius
  • Founder of the Society of Innovative Medicine

Jogundas Armaitis

  • PhD in Theoretical Physics, Utrecht University
  • 11 years of model building experience
  • Experience in applying Deep Learning to biometric and medical images
  • Marie Skłodowska-Curie Individual Fellow


Ignas Namajūnas

  • MSc in Computer Science, Vilnius University
  • Over 3 years of experience in computer vision industry and Deep Learning applications
  • Previously research lead for a Deep Learning based project
  • Two 1st places in Deep Learning competitions in Kaggle
  • 20th overall rank in Kaggle
  • Kaggle Grandmaster

Tomas Dirvonskas

  • BA in Physics, MA in Communications and Economics, Vilnius University
  • IT Business Competence, Royal Institute of Technology, Stockholm
  • 12+ years business experience within mobile product area
  • Successfully sold mobile product development agency
  • Organizer of AI conference and hackathon AI Camp

Get in touch

Arrange a Skype call or a visit to our office in Vilnius, Lithuania