Therapixel

07/08/2026

What happens when you add MammoScreen to a radiologist's daily workflow?

Same expertise. Same case volume. Better support. 🔍

We compared reading with and without AI side-by-side to highlight what really shifts in the reading room.

Swipe through ➡️ to see the breakdown, and let us know your thoughts in the comments!

06/08/2026

If you can't explain why your AI flagged a case, you can't use it in medicine.
Full stop.
The black box problem in medical AI isn't a technical curiosity. It's a clinical governance issue, a legal liability question, and a patient safety concern — all at once.

Here's the clinical reality:

A radiologist who acts on an AI recommendation is responsible for that decision. They cannot hand that responsibility to an algorithm and say "the model told me so." Their name is on the report. Their judgment is accountable.
Which means they need to be able to interrogate the AI's reasoning, not accept it blindly.

Here's the regulatory reality:
The EU AI Act explicitly requires explainability for high-risk AI systems. The FDA's evolving framework for AI/ML-based software as a medical device increasingly expects developers to demonstrate how and why their models produce the outputs they do. "It's a deep learning model and we can't fully explain it" is becoming less and less acceptable.

Here's what we do about it:
MammoScreen provides a Score from 1 to 10 - not a binary flag, but a graduated probability that a radiologist can weigh against their own assessment. It provides visual overlays showing the regions of the image that contributed most to the Score, so the radiologist can see what the model is responding to.

Is that perfect explainability? No. But it's a meaningful, clinically actionable level of transparency that supports rather than supplants the radiologist's judgment.
The explainability debate in medical AI is far from over. But any company not taking it seriously isn't ready for the clinical environment.

💬 What level of explainability would you require before trusting an AI tool in your reading room?

Read more about how MammoScreen works here: https://bit.ly/4q04pU4

Discover how MammoScreen works, using priors to deliver an actionable score to screening mammograms. Explore its seamless integration options.

03/08/2026

BEHIND THE ALGORITHM — PART 2 🧵
Most AI mammography tools read one image.
MammoScreen reads a conversation.
When an expert radiologist sits down to read a mammogram, they don't look at today's image in isolation. They pull up last year's. And the year before that. They look at how tissue has changed, where a density has shifted, whether something that wasn't there before is now appearing.
Image comparison is how expert mammography reading actually works.
So why do most AI tools ignore it?
Building an AI that meaningfully uses prior images is harder than it sounds. The model has to learn not just what breast tissue looks like, but what change looks like — specifically, the difference between normal variation over time and a change that's clinically significant.
That requires:
→ Training data that includes current and prior exams
→ A model architecture that can process multiple time points simultaneously
→ Calibration against cases where the prior comparison changed the diagnosis
MammoScreen was built to do all of this. It's the first AI to integrate prior mammography images directly into its cancer detection analysis.
The clinical result: a performance improvement that reflects how expert humans actually read — not how a single-image algorithm approaches reading.
💬 Has a comparison to a prior image ever changed a diagnosis for you? We'd like to hear.


Learn more about our Priors here:

Learn more about the unique way how MammoScreen analyses prior exams for more precise mammography readings

30/07/2026

Is AI coming for radiology jobs? Is it too complex to integrate? Does it slow down readings?

We hear these questions (and myths) all the time.

While AI is transforming breast cancer screening, there’s still plenty of misunderstanding about how it actually functions in the reading room.

We broke down the biggest radiology & AI myths in this quick carousel. 📊

👇 Check out the slides below and let us know your thoughts in the comments!

23/07/2026

"Is it approved?"
It's the first question anyone looking into integrating AI in their healthcare practice.
For MammoScreen, the answer is yes — twice over.

FDA 510(k) clearance (US):
Granted following a rigorous premarket review process. The FDA evaluates substantial equivalence to a predicate device — but for AI in radiology, that means demonstrating clinical safety and performance with real patient data.

CE marking (Europe, Class IIa):
MammoScreen holds a Class IIa medical device certification under the EU MDR — the classification for medium-risk devices with diagnostic function. This requires a notified body review of clinical evidence, quality management, and post-market surveillance planning.

What does this mean for you?

If you're evaluating AI for your radiology department, regulatory clearance tells you:
→ An independent authority has reviewed the clinical evidence
→ The developer has demonstrated safety and performance standards
→ There is a framework for monitoring the product post-deployment
Not all AI tools on the market have this. It matters — for patient safety, for clinical governance, and for your procurement due diligence.

We went through the hardest validation so you don't have to take a risk.

💬 Is regulatory clearance a key criterion in your AI evaluation process?

22/07/2026

Double reading in breast screening increases cancer detection rates up to 15% compared to single reading. Read that again.

The evidence is clear. The problem? There aren't enough radiologists.
In many countries, double reading is already the standard of care. In others, it's an aspiration the workforce can't sustain. Globally, we're heading toward a system where single reading becomes the norm — not by choice, but by necessity.

This is where the AI second reader debate becomes urgent.
The case for AI as second reader:
→ It's available for every case, without fatigue
→ It doesn't add to the radiologist workload — it supports it
→ The clinical evidence for AI-assisted reading is growing
→ Published studies (including our own) show sensitivity gains with AI support

The case for caution:
→ AI systems are not infallible
→ Regulatory frameworks for AI second reading vary by country
→ The "replacement" framing risks radiologist scepticism and adoption resistance
→ Not all AI tools are equal — evidence quality matters enormously

Our position: AI should be an evidence-based tool in the radiologist's hand, not a substitute for human expertise. The goal is better outcomes — not fewer radiologists. This is a debate worth having properly.

💬 Do you support AI-assisted second reading in your practice? What would it take for you to trust it? Let us know in the comments!

16/07/2026

Most women leave their mammogram not knowing whether they have dense breasts.
That needs to change.
Dense breast tissue does two things that matter clinically:
1. It makes tumours harder to see on a mammogram. Dense tissue and tumours both appear white — the technical term is "masking effect."
2. It increases cancer risk. Women with extremely dense breasts (BI-RADS D) have up to 4–6x the breast cancer risk of women with fatty tissue.
There are four categories (BI-RADS A–D). Around 40–50% of women fall into the dense categories (C or D).
In the US, federal law now requires that all mammography providers inform patients of their breast density. In Europe, guidance varies — but the clinical case is clear regardless of regulation.
MammoScreen includes automated density assessment as part of every read — so radiologists have the information they need, and patients have a basis for personalised conversations about their care.
Density isn't a footnote. It's a clinical variable that changes what screening looks like for nearly half of women.
💬 Should density notification be mandatory everywhere? Share your view.

15/07/2026

🚀 Le nouveau MammoScreen® est arrivé : Une IA de A à Z pour votre flux de lecture !
Nous sommes ravis d'annoncer l'obtention du marquage CE pour la toute nouvelle version de MammoScreen® ! Déjà plébiscitée par les radiologues français, notre IA franchit une nouvelle étape majeure.

Aujourd'hui, l'intelligence artificielle ne se limite plus à la simple détection de lésions. Cette nouvelle version a été conçue pour sécuriser chaque étape de votre pratique, optimiser vos flux de lecture et vous faire gagner un temps précieux grâce à 4 nouveautés clés :

🌟 4 innovations majeures au cœur de votre quotidien :
​​🛡️ L'évaluation de la qualité des images : Bénéficiez d'une évaluation automatique de la qualité technique des mammographies (basée sur des critères de positionnement, de détection du flou et de compression). Vous vous assurez ainsi que tous les examens lus sont de qualité optimale.
​​📊 L'analyse automatique de la densité : Obtenez une mesure objective et reproductible de la densité mammaire pour harmoniser et homogénéiser vos critères d'évaluation au sein de votre cabinet.
​​🔮 La prédiction du risque de cancer : Allez au-delà du diagnostic immédiat. L'IA évalue désormais le risque de cancer à 2 ans ou à 5 ans directement à partir de l’image, indépendamment de l’âge ou de la densité de la patiente. Un outil précieux pour individualiser les stratégies de dépistage, sans aucune charge de travail additionnelle.
​​✍️ Le pré-remplissage du compte-rendu : Le score, la densité et les critères de qualité sont automatiquement intégrés dans votre compte-rendu en seulement quelques clics. Recentrez votre expertise sur l'analyse, pas sur la saisie informatique.
🔌 Une intégration fluide et un déploiement simplifié
Aucun matériel supplémentaire n'est requis. Notre processus d'installation s'effectue entièrement à distance, en seulement quelques heures, pour s’adapter nativement à vos consoles et à vos habitudes de lecture (notamment via notre Widget en 1 clic).
Prêts à faire passer votre flux de lecture à la vitesse supérieure ?
👉 Découvrez le nouveau MammoScreen ici: https://www.mammoscreen.fr/decouvrez-le-nouveau-mammoscreen

Nous sommes ravis d'annoncer le marquage CE de la nouvelle version de MammoScreen. Celle-ci intègre de nouvelles fonctionnalités qui feront de cette IA en mammographie déjà la plus utilisée en France une ressource encore plus précieuse tout au long du flux de travail de dépistage du cancer du...

14/07/2026

🇫🇷 Joyeux 14 Juillet !
En cette Fête nationale, toute l’équipe de Therapixel célèbre l’innovation, l’excellence et l’esprit pionnier qui font la richesse de la France.

Nous sommes fiers de contribuer, chaque jour, à cette dynamique en développant des solutions d’intelligence artificielle au service de la santé, et en particulier de la détection précoce du cancer du sein.
Parce que la technologie peut (et doit) faire avancer la médecine, pour le bien de toutes et tous.

En ce jour de célébration, nous saluons aussi les femmes et les hommes engagés dans la recherche, le soin et l’innovation, en France et ailleurs.

Bonne fête nationale à toutes et tous ! 🇫🇷✨

10/07/2026

🌴 Next Stop: Orlando! Meet Therapixel at AHRA 2026!
The bags are packed, and the Therapixel team is heading to the Sunshine State! We are thrilled to be participating in the AHRA 2026 Annual Meeting this weekend at the Ga***rd Palms Resort & Convention Center in Orlando.

As medical imaging leaders gather to discuss the future of healthcare administration, operational efficiency, and team management, we are excited to showcase how advanced AI can transform your breast imaging workflow.

If you are looking to tackle radiologist burnout, streamline clinic throughput, and elevate imaging standards across your enterprise, stop by Booth #[Number] in the exhibit hall!

⚡ What We're Showcasing:
The MammoScreen® Widget: See how our "at-a-glance" native interface boosts reading ergonomics and enterprise productivity.

Automated Quality Check: Empower your technologists at the point of acquisition by catching positioning and compression issues in real-time, drastically reducing patient technical recalls.

AI Pre-Reporting: Save valuable time by automating data entry and populating AI findings directly into your reporting software.

Let's discuss how to successfully integrate and scale clinical breast AI across your entire healthcare network—with no hype, just results.

📍 Find Us in Orlando:

Event: AHRA 2026 Annual Meeting & Exposition

Location: Ga***rd Palms Resort & Convention Center | Booth #512

Dates: July 12–15, 2026

Attending the meeting this weekend? Drop a comment below or DM us to secure a dedicated 1-on-1 workflow deep dive with our experts on the show floor.

See you in Orlando! ☀️😎