Scribble supervised DSA sequence cerebrovascular segmentation with dual branch mutual learning

• 1、School of Artificial Intelligence, Beijing University of Posts and Telecommunications, 100876

Abstract：Cerebrovascular disease has long been a serious threat to the health of the world\'s people and is one of the major diseases facing the world today. Digital Subtraction Angiography (DSA), the "gold standard" for the diagnosis of cerebrovascular diseases, automatically segments cerebrovascular sequences in this mode, which is an important step for diagnosing related diseases and guiding neurointerventional procedures. Since the pixel-by-pixel annotation of blood vessels is time-consuming and labor-intensive, a simple graffiti annotation method can fully reduce the annotation difficulty. In the field of computer vision, the rapid development of deep learning methods has been widely used in the field of medical image segmentation. Based on this, this paper proposes a two-branch mutual learning segmentation method (MLDB) based on graffiti supervision, which embeds the main decoder and the auxiliary decoder with Dropout into the network, generates pseudo-labels using graffiti supervision on the resulting two sets of segmentation results, and subsequently relies on cross-pseudo-supervision and mutual consistency to constrain the network for consistency regularization. Experiments performed on the DSA sequential cerebrovascular dataset DIAS show that the method performs significantly better than state-of-the-art methods in recent years, achieving superior results.

Keywords： Computer Vision and Application Digital Subtraction Angiography Cerebrovascular Sequence Segmentation Weakly Supervised Segmentation