28 March 2016 in Journal of Clinical Investigation
Hemodynamic shear forces are intimately linked with cardiac development, during which trabeculae form a network of branching outgrowths from the myocardium. Mutations that alter Notch signaling also result in trabeculation defects. Here, we assessed whether shear stress modulates trabeculation to influence contractile function. Specifically, we acquired 4D (3D + time) images with light sheets by selective plane illumination microscopy (SPIM) for rapid scanning and deep axial penetration during zebrafish morphogenesis. Reduction of blood viscosity via gata1a morpholino oligonucleotides (MO) reduced shear stress, resulting in downregulation of Notch signaling and attenuation of trabeculation. Arrest of cardiomyocyte contraction either by troponin T type 2a (tnnt2a) MO or in weak atriumm58 (wea) mutants resulted in reduced shear stress and downregulation of Notch signaling and trabeculation. Integrating 4D SPIM imaging with synchronization algorithm demonstrated that coinjection of neuregulin1 mRNA with gata1 MO rescued trabeculation to restore contractile function in association with upregulation of Notch-related genes. Crossbreeding of Tg(flk:mCherry) fish, which allows visualization of the vascular system with the Tg(tp1:gfp) Notch reporter line, revealed that shear stress–mediated Notch activation localizes to the endocardium. Deleting endocardium via the clochesk4 mutants downregulated Notch signaling, resulting in nontrabeculated ventricle. Subjecting endothelial cells to pulsatile flow in the presence of the ADAM10 inhibitor corroborated shear stress–activated Notch signaling to modulate trabeculation.
25 November 2017 in arXiv
When supervising an object detector with weakly labeled data, most existing approaches are prone to trapping in the discriminative object parts, e.g., finding the face of a cat instead of the full body, due to lacking the supervision on the extent of full objects. To address this challenge, we incorporate object segmentation into the detector training, which guides the model to correctly localize the full objects. We propose the multiple instance curriculum learning (MICL) method, which injects curriculum learning (CL) into the multiple instance learning (MIL) framework. The MICL method starts by automatically picking the easy training examples, where the extent of the segmentation masks agree with detection bounding boxes. The training set is gradually expanded to include harder examples to train strong detectors that handle complex images. The proposed MICL method with segmentation in the loop outperforms the state-of-the-art weakly supervised object detectors by a substantial margin on the PASCAL VOC datasets.