FAQ

Frequently Asked Questions

General Questions

Therapixel™ is one of the leading French software companies specializing in artificial intelligence applied to medical imaging. MammoScreen™ is the latest of Therapixel’s products. Please go to Therapixel’s dedicated website to find more information about the company.

MammoScreen™ is commercially available through a SaaS (Software as a Service) model. You will:

  • Only pay for the amount of mammograms successfully processed
  • Always have the latest version of the software without any additional charge
  • Be able to scale up and down seamlessly and automatically as your organization evolves.

MammoScreen™ is available for sale in the US where it received a 510k clearance (reference K192854) from the FDA.

Yes, it is possible to test MammoScreen™ on your mammograms and/or directly in your practice.
Please send a message via our contact form by clicking here

We are always looking forward to receiving feedback from the people interested in our endeavor, as well as to getting help in accomplishing our objective, improving women’s health. Please send us a message via our contact form by clicking here.


Technical Questions

CAD stands for Computed Assisted Detection. As such, traditional CAD used for breast cancer screening are designed to detect abnormal findings. However, they can’t discriminate between benign and malignant patterns. Using machine learning, a subset of artificial intelligence (train a machine to learn from data) and the most advanced data science techniques (reproduce human intelligence from radiologists), MammoScreen™ can detect and also characterize the level of suspicion of findings. Consequently, MammoScreen™ can display only the suspicious findings on the interface and don’t overwhelm the work space with benign findings. MammoScreen™ provides actionable information to support your interpretation of the exam by giving you a level of suspicion at the mammogram, breast and lesion levels.

In the US, MammoScreen™ is intended for use as a concurrent reading aid for interpreting physicians, to help identify findings on screening FFDM acquired with compatible mammography systems and assess their level of suspicion. Output of the device includes marks placed on findings on the mammogram and level of suspicion scores. The findings could be soft tissue lesions or calcifications. The level of suspicion score is expressed at the finding level, for each breast and overall for the mammogram. Patient management decisions should not be made solely on the basis of analysis by MammoScreen™.

MammoScreen™ is intended to be used with screening FFDM acquired with compatible mammography systems.

One should not use MammoScreen™ information (even if a report is generated on the interface) to guide clinical decision for:

  • poor quality images (e.g. blurry images, bad acquisition, etc.)
  • non-standard mammographic views (e.g. projections, magnifications, compressions, etc.)
  • synthetic 2D images
  • women with previous breast surgery or breast implants or internal breast markers
  • women with a pacemaker

Remember that MammoScreen™ only uses the mammogram, and no additional information available to the interpreting physician. As a consequence, some detected findings may not be relevant for the user (False Positive). Some soft tissue lesions or calcifications may also be missed or undervalued (False Negative).

MammoScreen’s users should refer to the User Manual for a complete list of indications of use.

MammoScreen™ detects all the potential cancer lesions and displays differently t2 subgroups:

  • Soft tissue lesions including mass, asymmetries, distortions, and lymph nodes
  • Cluster of calcifications

The soft tissue lesions are represented by a circle and the calcifications are represented by a triangle.


Integration in your practice

Processing time is very quick and mostly depends on your network speed to send the mammogram to our cloud and receive the report. Usually it takes less than 2 minutes overall and is used in centers where mammograms are interpreted immediately after acquisition.

MammoScreen™ has not been designed to be integrated into your mammography reader. Placed on a separate screen, it enables each radiologist to look at the AI information when he/she needs it during the interpretation process. If the display as an overlay in your viewer is a prerequisite for you, please contact us via our contact form by clicking here to discuss feasibility.

Yes. Depending on your IT infrastructure, we can easily deploy at one or multiple sites. Please contact us to together define the best IT strategy. Please send us a message via our contact form by clicking here and the right person will come back to you.

We are HIPAA compliant and we are a HIPAA compliant cloud host to ensure maximal protection. Moreover we systematically de-identify every DICOM image leaving your practice and systematically delete images from our cloud as soon as the output report has been successfully received (a few minutes). If you have any questions, don’t hesitate to ask our Data Protection Officer (DPO) directly by sending an email to “dpo @ therapixel.com”


Underlying AI technology & Clinical validation

MammoScreen™ is a concurrent read software and as such had to demonstrate an increase of performances of radiologists when using the software while interpreting cases in order to receive FDA clearance. Standalone performances can’t be directly translated into an increase of performance of radiologists.

In a Multi-Reader, Multi-Case study, we have shown that MammoScreen™ improves radiologists cancer detection rate by 5 percentage points while lowering their false positive rate by 6 percentage points. The paper of this study is currently in press on Radiology: Artificial Intelligence, more details to come soon.

Our database is made  of over 1 million screening mammograms coming from 15 sites both in the EU and the USA, and from both private and public health institutions. The diversity of sources ensures a fair representation of populations of all origins living in these countries (Caucasian, African American, Hispanic, etc.). It is also largely enriched with cancers to cover a multitude of different lesion types (invasive, in situ, masse, calcification, asymmetry, distortion). Among these cases, there is a natural distribution of density type for these  population.

We have tested our algorithms on 2 different data sets never used by the algorithm for training (one US, one EU). This dataset of screening mammograms is composed of 1,700 cancers (positive biopsy within 3 months of the screening exam) and 20,000 true negatives (confirmed negative on the next screening, no cancer found in the life of the woman). Note that only one exam per patient was used.