🔬 Unlocking the Power of AI-based Image Analysis — Part III 🔬
The last two episodes demonstrated AI's ability to handle complex segmentation tasks, both in fluorescence and color images. However, existing AI models are sometimes limited in their ability to handle unforeseen variations (outlier, poor image quality, change in imaging setup). Fortunately, there's a solution!
The transfer learning 🖥️
In fact, it is possible to manually add relevant labelled data to the database to re-train a model and help it improve performance. At Imactiv-3D, we have developed an open-source segmentation Napari plugin called Sketchpose, enabling us to offer transfer learning on any type of data to improve results and seamlessly achieve better segmentation.
Bacteria counting 🦠
In today’s illustration (1), the image is a phase contrast microscope acquisition of bacteria. Segmentation of contiguous filiform objects such as bacteria is often difficult, and most existing models do not directly lead to perfect segmentation results that would enable reliable characterization. Next is a first unsatisfactory segmentation result with an existing model (2) then improved using our transfer learning method (3), leading to a much better output.
Stay tuned! 📢
Stay tuned for further insights into how our expertise in various imaging techniques can improve scientific analysis.
#Imactiv3D #ImageAnalysis #Segmentation #Research #AI #Napari