PYCAD

Did you know that you can build a fully functional MVP for medical imaging using Python and Trame? Although Trame isn't widely known in the community, its capabilities are impressive. It allows you to integrate all the tools and features of VTK into a stable and functional web application. However, like any tool, it has its limitations. In my experience, one major drawback is the documentation, which is quite weak. You often need to search extensively to find even small explanations for specific tasks. Overall, working with Trame has been a good experience, though it can be challenging at times. If anyone has already used Trame to build an MVP, I’d love to hear about your experience! #trame #vtk #medicalimaging #healthcare #nnunet #visualization

I think this turned out better than I expected! When we planned to create a course on how to use nnUNet, our goal was simply to help people get started. We weren't focused on achieving the highest possible model accuracy since that wasn't our main objective. However, as always, nnUNet has a way of surprising you. Even when you're not aiming for the most accurate model, it delivers impressive results. Haha! We're excited to share our latest course with you. It provides a step-by-step guide on how to train an nnUNet model and how to evaluate it. Once you have your model, you might want to explore where you can deploy it, whether for your supervisor or for investors. At the end of the course, we have a demo section showcasing three deployment options: Streamlit + Pyvista, PyQt + VTK, and Trame + VTK. BTW: the segmentation shown in the video in this course, is the output of an evaluation case from the model! Learn more about the course using the links below: YouTube: https://lnkd.in/edtxJhaU Code: https://lnkd.in/eUkZGQmK Dataset: https://lnkd.in/eUGrkCNf Interested in what we do in the medical imaging field? Subscribe to our newsletter for weekly updates 🚀: pycad.co/join-us

I'm doing some experimentation with Trame. Here's a quick explanation of what you're seeing below ↓ This is a web application built purely in Python. It's using a library called Trame. That thing that looks like an alien is the Aorta. Which is the largest blood artery in the human body! This is actually the output of a deep learning model that we trained to segment the aorta automatically. The output of this model is a 3D array that we saved as Nrrd file for easy manipulation. We then reconstructed an STL object using VTK library. Then we are rendering the STL object directly on the browser using Trame. All of this process is automated! The output looks very slick if you ask me 😂 This is the first time I try Trame, and I have to say that I'm impressed so far! I will probably document more about my experiments with it, so stay tuned! At PYCAD we have built several models like this for our clients, for automatic segmentation of different organs and anatomical structures. We have also built web apps for them that allow them to easily run the AI models in the backend and do visualizations like this in the frontend (using other tools, not Trame). Interested? DM me! #automaticsegmentation #medicalimaging #machinelearning #trame

Some people ask me what we use to annotate medical imaging data. Here's my answer ↓ Usually, we don't annotate medical data by ourselves. We work with certified 3rd party companies to do this. But what if I want to do some minor modifications on some annotation? In this case, we use 2 freely available software programs: 3D Slicer and ITK Snap. They both make it easy to visualize all sorts of medical data formats including: DICOM, NIFTI and NRRD. Btw, we don't just use them to make modifications to medical annotations. We also use them to do data verification and validation, which is a crucial step in every project we take at PYCAD. When do we use which? Generally speaking, ITK Snap is my go to for a quick data visualization and checking. Mostly because it opens up quicker than 3D Slicer 😅 But if I want to do a thorough check of the data and have more control, then I go with 3D Slicer. But when I want to do a thorough data verification and validation, I actually open the file I'm inspecting in both softwares. This helps in some edge cases where the person who did the annotation used 3D Slicer and saved the file in some generic way. When an annotator does this, it could hide some information that are crucial for any ML project. For example, we noticed that sometimes when we open an annotation file in 3D Slicer, we see: segment_1, segment_2, ... These would be the names of the annotated classes. If you just look at this, you'll think that segment_1 has label 1, segment_2 has label 2 so on.. But in reality, it could be that segment_1 has label 2, segment_2 has label 5, ... This could happen when the annotator creates a segment (annotation), then deletes it and creates a new one. 3D Slicer keeps the information of the deleted segment but hides it away. When we open the same patient file in ITK Snap, we immediately notice the discrepancy in the labels. This is just one of the possible scenarios. Keep this in mind next time you're working on a medical imaging project. 💊 Btw, if you need help developing AI solutions for medical imaging, we at PYCAD can help you with that. Feel free to DM for more information! ➡ And if you like this kind of medical imaging content, then you'll love our weekly newsletter. You'll get some freebies right after you join! Join here: newsletter . pycad . co

To run our business efficiently at PYCAD we keep extensive documentation on Notion. We try to document most of the technical work we do. This helps us go fast when building medical imaging solutions for our clients. A new feature from Notion is going to help us go even faster. Now, you can ask Notion to find answers to questions related to your workspace. For example, we have a documentation about how to use nnUNet. So I asked Notion to tell me how I can evaluate a nnUNet model. Not only did it give me the exact answer I needed. It also, gave me links to my Notion pages that contain that information! The integration of LLMs in every day workflows is a game a changer!

New model added to the list! Look how clean these lungs are segmented 🚀

Check out our new medical imaging app! Upload your CT scans as DICOMs and get the segmentations of the liver, spleen, heart, spine, and hips (pelvis) as NIFTI files! You can use the outputs to train your own models! This is what we call "model-led annotation". The application will be 100% free of charge and help you accelerate your annotation process. Stay tuned for the release date! #medicalimaging #machinelearning #dicoms #modelledannotation