Identifying and relocating individual cells, especially sperm for IVF, is challenging due to their similar look, rotation, and transparency. We propose a new method using a vision transformer network, trained on 3D images of cells taken with a holotomographic microscope. Our approach uses a Detection Transformer (DETR) network, which makes unique predictions about the cells and their context. This method improves the speed and accuracy of cell identification in IVF, offering potential for better success rates and optimization of high-throughput imaging.
|