CoLumbo is a software for analysis and diagnosis of lumbar spine images obtained with MRI.  The purpose of CoLumbo is to assist with the reading of MRI images on the lumbar spine and to provide information on detected pathologies and abnormalities, and to reduce the time to record the readings.
A few of the most common pathologies of the spine are: disc herniation, stenosis, spondylolisthesis and lordosis. They are also among the most common causes of spine surgery. Evidence indicates that lower back pain is a global and extremely serious problem that can interfere with normal life. 

Value Proposition

The key novelty of CoLumbo lies in the unique machine learning algorithm, based on convolutional neural networks combined with medical domain knowledge, dropout regularization and packet normalization which are capable of reading MRI images, in order to provide a complete diagnosis. Unet-like architecture is used to improve the resolution.
The first step is the Segmentation of specific structures which is made on a pixel level and the use of fully convolutional networks allows quick segmentation maps to be generated for images of any size. Second is the identification of key points in the MRI images. After this is the measurement of distances, areas and angles. Finally the combination of segmentation, measurements and specially applied algorithms provides the diagnosis according to the parameters described in the scientific medical literature.

Intended Use

Modality - MR; Body part - Lumbar spine (L-Spine); Magnetic Field Strength 1,5-3T; Age - 18 and above;
At least one sagittal T2w sequence containing 5 slices or more.
MR Acquisition Type - 2D or 3D; Scanning Sequence - SE; Repetition Time: at least 1040 ms; Echo Time: from 87.8 ms to 121.0 ms
Maximum slice thickness 5mm; Min Sagittal Resolutions (Cols x Rows): 384x512
At least one axial T2w sequence; Min Axial Resolutions (Cols x Rows) 276x192; Maximum slice thickness 5mm; MR Acquisition Type - 2D or 3D ; Scanning Sequence - SE; Repetition Time: at least 1040 ms Echo Time: from 96.1 ms to 111.0 ms


Not intended to analyze images of pregnant persons. Do not support protocols with contrast injections (even if there are series w/o contrast it usually contains fracture, infection, postoperative complications or tumor that we will support in the next version.

Model performance metrics

Everything that we have is in 'Performance or ROC curves figures.pdf'

Performance Curves

Performance Curves

Performance Curves

Performance Curves

Performance Curves

Performance Curves

The test device performs diagnostics with characteristics not worse than those specified in Table 1a.
The accuracy of the segmentation algorithm (in% of measured key points with deviation below the threshold of clinical significance) is not worse than specified in Table 1b.
The accuracy of the measurement algorithm (in% of measurements with deviations below clinical significance according to the investigator) using the test device is not worse than that specified below in Table 1c, for the maintained series and sections.
The difference in the degree of agreement with the investigator arbitrator of the radiologist using the software compared to the radiologist not using the software in the disputed cases is not less than the difference indicated in Table 1d.
( i Presence of hernia is noted - yes / no; degree of hernia - light, medium, severe; position of hernia in the presence - median, left / right paramedian, left / right foraminal; left / right extraforaminal.)
The accuracy of the algorithm for automatic completion of the diagnosis modifiers in the cases of visually accurate segmentation is not worse than the correct column shown in Table 1e. The modifier determined by the software is considered accurate if it coincides with the opinion of a research radiologist.


Smart Soft ltd.

SmartSoft is founded in 2004 and is developing products for Optical
Character Recognition (OCR); data capture solutions and image processing
and provide quick, precise, and reliable image or document reading and analysis.
The team working on CoLumbo is international with vast experience in project development, finance, technology and medicine. Consultants to the project are radiologists, neurosurgeons and MSK specialists from the EU and USA.

Learn more