About

Segmentation in radiology

Segmentation of radiological images is important in many fields. Volumetry, visualization including VR/AR, 3D printing, radiotherapy, (co-)registration, and many other post-processing tools are some of the examples that require segmentation.

Meanwhile, segmentation has traditionally been regarded as laborious and uninteresting. However recent progress in the field of computer vision has enabled the use of AI for both automated and interactive segmentation. For radiology, we hope that these new developments will hasten the implementation of broadly used volumetry instead of the subjective ruler-based distance measurements that are still the current clinical standard.

About us

We are two radiologists based in Oslo, Norway who crave more AI-enhancement in daily clinical practice. As hobby-programmers, we are having fun and learning a lot while building MedSeg. We have reached the level where we find this tool useful for effectively creating segmentations of high quality. Hopefully, as MedSeg continues to improve, it will become useful to other people too.

MedSeg is based on the following openly available tools and resources:

DICOM reader: https://github.com/rii-mango/Daikon

Nifti reader: https://github.com/rii-mango/NIFTI-Reader-JS

Deep learning (DL) in browser: https://www.tensorflow.org/js

Development of DL models: https://keras.io/ + https://www.tensorflow.org/

DeepGrow module: https://arxiv.org/abs/1903.08205

Ideas and experience from using another, Python-based application with AI-capabilities. RILContour: https://link.springer.com/article/10.1007/s10278-019-00232-0

Another great advanced and open-source application filled with loads of post-processing tools, 3D Slicer: https://www.slicer.org/

Overview of web-based DICOM viewers: https://medevel.com/14-best-browser-web-based-dicom-viewers-projects/