eduzhai > Applied Sciences > Engineering >

Faster Mean-shift GPU-accelerated clustering for cosine embedding-based cell segmentation and tracking

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 15 pages

Abstract: Recently, single-stage embedding based deep learning algorithms gainincreasing attention in cell segmentation and tracking. Compared with thetraditional "segment-then-associate " two-stage approach, a single-stagealgorithm not only simultaneously achieves consistent instance cellsegmentation and tracking but also gains superior performance whendistinguishing ambiguous pixels on boundaries and overlaps. However, thedeployment of an embedding based algorithm is restricted by slow inferencespeed (e.g., around 1-2 mins per frame). In this study, we propose a novelFaster Mean-shift algorithm, which tackles the computational bottleneck ofembedding based cell segmentation and tracking. Different from previousGPU-accelerated fast mean-shift algorithms, a new online seed optimizationpolicy (OSOP) is introduced to adaptively determine the minimal number ofseeds, accelerate computation, and save GPU memory. With both embeddingsimulation and empirical validation via the four cohorts from the ISBI celltracking challenge, the proposed Faster Mean-shift algorithm achieved 7-10times speedup compared to the state-of-the-art embedding based cell instancesegmentation and tracking algorithm. Our Faster Mean-shift algorithm alsoachieved the highest computational speed compared to other GPU benchmarks withoptimized memory consumption. The Faster Mean-shift is a plug-and-play model,which can be employed on other pixel embedding based clustering inference formedical image analysis. (Plug-and-play model is publicly available:this https URL)

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×