MedSAM: A New Tool for Medical Image Segmentation

<p>😷 Recently, I’ve been thinking about how machine learning models can change the medical industry, especially in the field of detection and segmentation. A group of researchers is already developing MedSAM - a universal tool for segmenting medical images.</p>
<p>MedSAM is a modification of the Segment Anything Model (SAM) (the one from Meta, which I wrote about earlier), which was previously used for segmenting any images. However, applying SAM to medical images turned out to be not so effective. The researchers compiled a massive dataset consisting of over 200,000 masks representing 11 different modalities of medical images. They then developed methods to adapt SAM to the tasks of medical image segmentation.</p>
<p>As a result, MedSAM demonstrated a significant performance improvement compared to the original SAM model. On average, the model showed an increase in the Dice Similarity Coefficient (DSC) by 22.5% and 17.6% for 3D and 2D segmentation tasks, respectively.</p>
<p>It’s truly amazing how all industries, including medicine, are changing.</p>
<p>Paper: <a href="https://arxiv.org/abs/2304.12306">https://arxiv.org/abs/2304.12306</a><br>Github: <a href="https://github.com/bowang-lab/MedSAM">https://github.com/bowang-lab/MedSAM</a></p>;