Smart Soft performed some tests that will demonstrate that, in general, CoLumbo performs segmentation of C-Spine exams and even detects some types of pathology. So the next version will be ready in 2022 and will support C-spine. In 2024, we will support the CT modality as well.

Smart Soft intends to include Foraminal stenosis, Modic changes, Schmorl’s nodes, Scoliosis, Fractures, Osteophytes, Disc degeneration disease,HIZ, Sequestration, Subarticular stenosis and Ligamentum Flavum Hypertrophy in the next version will reach and possibly surpass the human level. However, some very rare types of pathology will not be included soon. Smart Soft’s product is an assistant tool for physicians and not a product that replaces physicians. The physician may agree to the CoLumbo report but may also choose to add or remove some texts.

Segmentation is fully automatic. In some cases, the user might change the key points and as a consequence, some measurements will be automatically performed again.

All measurements are calculated; however, most are not usually included in the radiologists’ report. For that reason, the release version includes the measurements correlated to the detected pathologies. Other measurements, like foramen size and spinal canal diameter, are also included. This is because they can help confirm detected pathologies or identify others that are undetected.

Yes. The full report can be saved as .rtf or .pdf or sent to a printer. It is also part of the entire exported study in .pdf format.

The current version of CoLumbo (V1) provides 70% – 80% sensitivity and 90% specificity.

Technically, we can propose different packages to customers, including price per exam, per year, and unlimited. That way, we can follow the pricing schema of our partners. The basis of our pricing is that we save at least 25% of time costs. This is 25% of 50 USD in Western countries, which means 12.5 USD. Consequently, this equates to 3,500 -12,000 USD in annual savings.

Our first release version is already CE mark certified and FDA 510(K) cleared, making CoLumbo the first spine AI solution for MRI modality with regulatory clearance in both Europe and the USA.

Concerning the number of features – this means more time and effort for risk management, usability engineering, testing, and clinical investigations, but we do not believe certification will be delayed. Clinical trials were performed in 3 hospitals and finished successfully in October 2020.

Yes, we can provide API to easily integrate the software into 3rd party software. Our standalone version will be mainly for demonstration purposes so one can download the exam from the PACS and, after that, load it into CoLumbo. The user can even change the settings of the PACS to allow for automatic download and upload of studies.

We may deploy software on their server, PC or the working station provided minimum requirements are met.

The software has minimum requirements, that must be met, for it to run. Depending on the deployment methods, there are differences, but there are general requisites for both: the machine should have at least 8GB of RAM, 600 MB free space for the installer and 4 GB for the installed software, 2.10 GHz CPU with at least 2 cores, Windows 10 or Windows 7 as an operating system.
Depending on the deployment, these additional requirements have to be met:

Client-Server: Windows Server 2016 operating system, installed on the server side, an NVidia GPU with 2 GB memory and CUDA installed (exactly version 11.0), internet connection with minimum speed of 4 MBs;

Standalone: a local NVidia GPU with 2GB memory and CUDА installed (exactly version 11.0);

Version with study samples: 5 GB of additional free space.

Windows 10, Windows 7, or Windows Server 2016. The API can be used on Linux, too and it is possible to be used in specialized hardware. For predictions on a GPU, CUDA version 11.0 is required.

General:

• Modality – MRI;
• Body part – Lumbar spine (L-Spine);
• Age USA – above 18 years old;
• Age EU – between 18 and 70 years old;

For sagittal:

• At least one sagittal T2w sequence containing 5 slices or more.
• MR Acquisition Type – 2D or 3D;
• Scanning Sequence – SE;
• Repetition Time: at least 940 ms;
• Echo Time: from 60 ms to 300 ms;
• Maximum slice thickness 5mm;
• Min Sagittal Resolutions (Cols x Rows): 280×280;
• Maximum voxel volume – 1.9 mm2 for 2D series and 1.0 mm3 for 3D series.

For axial:

• At least one axial T2w sequence.
• Min Axial Resolutions (Cols x Rows) 192×192;
• Maximum slice thickness 5mm;
• MR Acquisition Type – 2D;
• Scanning Sequence – SE;
• Repetition Time: at least 940 ms;
• Echo Time: from 60 ms to 300 ms;
• Maximum voxel volume – 1.6 mm2.

Not intended to analyze images of pregnant patients or patients with prior surgeries. Does not support protocols with contrast injections (even if there are series without contrast it usually contains fractures, infections, postoperative complications or tumors that we will support in the next version). Does not support distorted or blurry images. Does not support patients with scoliosis.

For fat suppression and other sequences different from T2 spin-echo, the algorithm can support them both as single or additional sequences but only for research purposes. These capabilities will not be tested during the official clinical trials. We need two weeks per new sequence, as with the Dixon protocol we covered for the MICCAI challenge.

-ResNet for feature map extraction.
-UNet for segmentation.
-3D convolution for 3D segmentation refinement.

We will not ask for a change in the protocols used. If the protocol contains a T2 spin-echo sequence, everything is OK. Protocols that do not contain a T2 spin-echo sequence are usually performed with some specific pathology in mind that is not among the most common that we support. For the next version, we will add support for 20 more pathologies and some of them will require supporting other protocols. Of course, for some special-purpose protocols, we will show a warning that the protocol is not currently supported.

Absolutely. The user can define the positive probability of pathology so he/she can work on high sensitivity mode. These can be different for different pathologies so one can play that way with the sensitivity vs. specificity tradeoff.

In the current version, no. The next version will need T1 for Modic changes and accuracy improvements. However, the user can still load the series manually and review it.

Yes, if costal facets, and bifurcation of the aorta all indicate the same labeling; otherwise, the full spine image is needed and we just indicate that there is a transitional vertebra and labeling is uncertain.

Yes, pathology can be added or removed with one or two clicks. Also, the user may add pathologies, unsupported by the software, as well as free text. The text of the report of the verified findings can also be edited.

Yes, the user can see the color-coding of the level labels and quickly navigate to the chosen level by selecting it from the level dropdown menu. The user can also navigate to the most prominent slice of a certain pathology by hovering over it in the pathology panel.

The software is designed as an assistant type of software. The first version saves 25% of the time. The aim is for the next versions to save up to 80-90% of the time. However, the radiologists will be needed for at least the next 40 years.