info@ira-labs.com
Ru En

AI devoted to the radiologist

CHEST-IRA is a service for automatic complex chest CT analysis using state-of-the-art artificial intelligence technologies
To date, IRA Labs analyzed 

1012143
scans
Lung
window
Soft tissue
window
Bone
window

Our vision and values

The number of scans increases every year, leading to the radiologists' burden growth, thus affecting the quality of diagnosis.

IRA Labs enhances the work environment 

and quality of diagnostics

01

Minimize the risk of the errors,

comprehensively analyzing the entire scan, and focusing on both primary pathology and comorbidities.
02

Reduce analysis time,

performing all routine calculations automatically.

Our solution

Identification and visualization of pathologies

Automated scoring and classification

Generation of radiology reports

Intuitive
interface

Chest-IRA -
chest CT scans complex analysis

Respiratory system Cardio-vascular system Musculo-skeletal system Abdomen
COVID-19

COVID-IRA module assesses CT findings suspicious for COVID 19: ground-glass opacity, consolidation, and reticular changes.

Accuracy: ROC AUC = 0.98

Lung cancer

LungNodules-IRA module searches for lung nodules to identify malignant neoplasm (C34) with linear size and volume measurement.

Accuracy: ROC AUC = 0.93

Emphysema, honeycomb lung

Emphysema detection (J43).

Development phase


Pneumothorax, hydrothorax

Detection of pneumothorax, hydrothorax, hemothorax, pyothorax (J90-J94).

Accuracy: ROC AUC = 0.99


Lymph node enlargement

Detection of mediastinal lymph nodes enlargement (C77,C59).

Development phase


Coronary artery calcification

Agatston-IRA module detects the signs of coronary artery calcification (I25) and quantifies using the Agatston score.

Accuracy: ROC AUC = 0.98


Aortic aneurysms and dilatations

Aorta-IRA module measures the diameters of the ascending and descending aorta to detect aneurysms and dilatations (I79).

Accuracy: ROC AUC = 0.99

Pulmonary artery dilation

Pulm Trunk-IRA module measures pulmonary trunk diameter to detect pulmonary hypertension (I25).

Accuracy: ROC AUC = 1.0


Pericardial and epicardial fat

CardiacFat-IRA module measures the amount of adipose tissue surrounding the heart (pericardial and epicardial fat) (I27).

Accuracy: ROC AUC = 0.99

Osteoporosis and vertebral compression fractures

Genant-IRA module assesses vertebral body mineral density for signs of osteoporosis (M81), measures vertebral body deformity, and performs a Genant classification for compression fractures.

Accuracy: ROC AUC = 0.99

Sarcopenia

Evaluation of muscle and adipose tissue for the detection of sarcopenia (M62.5).

Development phase

Rib fractures

Automatic detection of rib fractures.

Development phase

Adrenal masses

Assessment of the adrenal glands for signs of neoplasms (D44). When finding a mass, its linear size and volume are measured.

Version 1.0

Fatty liver disease

Assessment of liver density to detect fatty infiltration (K70-K77).

Development phase

Liver tumors

Liver evaluation for benign and malignant masses (C22).

Development phase

* ROC AUC is a statistic that allows you to evaluate the quality of a binary classification. The closer the indicator is to one, the better the classifier.

* ROC AUC is a statistic that allows you to evaluate the quality of a binary classification. The closer the indicator is to one, the better the classifier.

* ROC AUC is a statistic that allows you to evaluate the quality of a binary classification. The closer the indicator is to one, the better the classifier.

* ROC AUC is a statistic that allows you to evaluate the quality of a binary classification. The closer the indicator is to one, the better the classifier.

Why IRA Labs?

Complex approach

We analyze all the organs presented in the scan, which helps the radiologist and clinician get a complete picture of the patient's condition without missing anything, cope with the routine faster, and focus at once on differential diagnosis.

High-quality algorithms

A strong and dedicated development team of algorithmic scientists, consisting of graduates of the Moscow Institute of Physics and Technology and Skoltech. A team of expert radiologists participates in the training of each algorithm. All modules go through several stages of testing and are constantly being improved.

Leader in Moscow AI experiment

Since 2020 Moscow's healthcare departments have experimented with innovative computer vision technology for medical image analysis with the participation of 21 companies (including international). Currently, IRA Labs is the leader in terms of quality and quantity of products:
  • COVID-19 - ROC AUC 0.98
  • Pulmonary nodes - ROC AUC 0.93
  • Aortic aneurysms - ROC AUC 0.99
  • Coronary artery calcification - ROC AUC 0.98
  • Pericardial/epicardial fat - ROC AUC 0.99
  • Osteoporosis - ROC AUC 0.99

Compliance with clinical practice guidelines

Our product speaks the same language as a doctor: we build an interpretable result following international clinical guidelines for each finding.
At IRA LABs, we are deeply committed to advancing technology while upholding our responsibility to the environment. In our latest initiative, we've successfully optimized our AI screening system to significantly reduce its computing power requirements. This strategic enhancement not only bolsters system efficiency but also substantially lowers our carbon footprint. By integrating these sustainable practices, we're not just improving our operations; we're contributing to a healthier planet for future generations.

Advanced scientific results

The product is built based on cutting-edge scientific results obtained in collaboration with the Skolkovo Institute of Science and Technology.

1 / 0

M. Goncharov, M. Pisov, A. Shevtsov, B. Shirokikh, …, M. Bel...

Medical image analysis 71, 102054

We have proposed a novel computer vision algorithm that improves patient triage by simultaneously identifying symptoms of COVID-19 and assessing severity. The paper was published in Medical Image Analysis, the leading radiology journal.

M. Pisov, V. Kondratenko, A. Zakharov, …, M. Belyaev

2020, MICCAI

We have proposed the world's first fully automatic algorithm for assessing the height of the vertebral bodies and detecting compressions following the recommendations of the International Osteoporosis Association.

B. Shirokikh, A. Shevtsov, A. Dalechina, E. Krivov, …, M. Be...

2021, Journal of Imaging
Исследование алгоритмов детекции и сегментации новообразований различной локализации. Был предложен специализированный алгоритм, снижающей время обработки исследований до 40 раз.

B. Shirokikh, I. Zakazov, A. Chernyavskiy, I. Fedulova, M. Bel...

2020, Domain Adaptation and Representation Transfer, and Distri...
Study of the effect of changing scanning protocols on the operation of segmentation neural networks. A simple but effective way of retraining on data from protocols new to the algorithm is proposed.

B. Shirokikh, A. Shevtsov, A. Kurmukov, A. Dalechina, …, M. ...

2020, MICCAI, top-1 conference in the field
Systematic study of the low-quality problem in the detection of small neoplasms. A new method that increases the sensitivity of determining such tumors by 3-5 times has been proposed.

B. Shirokikh, I. Zakazov, A. Chernyavskiy, M. Belyaev

2021, MICCAI, oral talk
A comprehensive study of the effect of changing scan protocols on the internal parameters of deep convolutional networks has been carried out. A new retraining method adapting to a different amount of data with new, previously unknown protocols is proposed.

A. Petraikin, Z. Belaya, A. Kiseleva, Z. Artyukova, M. Belyaev,

2020, Problems of Endocrinology
Clinical study of the algorithm for assessing the vertebral bodies' heights to detect signs of osteoporosis. The result ROC AUC was 0.95-0.98.

V. Chernina, M. Pisov, M. Belyaev, I. Bekk, K. Zamyatina, T. Ko...

2020, Kardiologiia
Clinical study of the algorithm for assessing the epicardial fat volume. We showed that automation reduces the analysis time by 30 times at a correlation of 0.95 compared with manual measurement.

«We strive to provide the best quality of algorithms and doctors' convenience to benefit the highest quality patient diagnostics»

Victor Gombolevskiy, IRA Labs CBDO

We are IRA Labs

Our goal

To create a high-quality service that will become an indispensable assistant to a radiologist. We strive for artificial intelligence to take over routine measurements, help identify the underlying disease and comorbidities, to ensure the highest quality of diagnosis.

Our story
Since 2016, our research team has studied machine learning methods in biomedical problems. In 2020, when the pandemic dramatically increased the flow of CT scans, we saw an opportunity to apply our knowledge to automated CT analysis tasks. We were one of the first to make a product for COVID-19. Our solution was recognized as the leader among 15 competing companies and had the highest ROC AUC value among publicly available services.
Our name
Our values are reflected in the name: IRA Labs is short for Intelligent Radiology Assistance Laboratories.

A Passionate Team

Victor Gombolevskiy
CBDO
Radiologist, PhD.
Maria Dugova
CMO

Radiologist, 13+ years experience

CT and MRI expert, member of the European Society of Radiology (ESR)

Regina Gareeva
Product Manager
MS (MIPT)

4 years of analytical work and product management in healthcare organizations
Ekaterina Petrash
Radiologist

Radiologist, PhD, 11+ years experience

CT and MRI expert, member of the European Society of Radiology (ESR)

Join the IRA Labs team

Send your CV to join@ira-labs.com

Ru En
Contact us

Contact us