A deep learning-based strategy for producing dense 3D segmentations from sparsely annotated 2D images.

Producing dense 3D reconstructions from biological imaging data is a challenging instance segmentation task that requires significant ground-truth training data for effective and accurate deep learning-based models. Generating training data requires intense human effort to annotate each instance of an object across serial section images. Our focus is on the especially complicated brain neuropil, comprising an extensive interdigitation of dendritic, axonal, and glial processes visualized through serial section electron microscopy. We developed a novel deep learning-based method to generate dense 3D segmentations rapidly from sparse 2D annotations of a few objects on single sections. Models trained on the rapidly generated segmentations achieved similar accuracy as those trained on expert dense ground-truth annotations. Human time to generate annotations was reduced by three orders of magnitude and could be produced by non-expert annotators. This capability will democratize generation of training data for large image volumes needed to achieve brain circuits and measures of circuit strengths.

Deep inspiratory breath-hold radiotherapy on a Helical Tomotherapy unit: Workflow and early outcomes in patients with left-sided breast cancer.

Introduction: The clinical implementation of deep inspiratory breath-hold (DIBH) radiotherapy to reduce cardiac exposure in patients with left-sided breast cancer is challenging with helical tomotherapy(HT) and has received little attention. We describe our novel approach to DIBH irradiation in HT using a specially designed frame and manual gating, and compare cardiac substructure doses with the free-breathing (FB) technique. Material and methods: The workflow incorporates staggered junctions and a frame that provides tactile feedback to the patient and monitoring for manual cut-off. The treatment parameters and clinical outcome of 20 patients with left-sided breast cancer who have undergone DIBH radiotherapy as a part of an ongoing prospective registry are reported. All patients underwent CT scans in Free Breathing (FB) and DIBH using the in-house Respiframe, which incorporates a tactile feedback-based system with an indicator pencil. Plans compared target coverage, cardiac doses, synchronizing treatment with breath-hold and avoiding junction repetition. MVCT scans are used for patient alignment. Results: The mean dose (Dmean) to the heart was reduced by an average of 34 % in DIBH-HT compared to FB-HT plans (3.8 Gy vs 5.7 Gy). Similarly, 32 % and 67.8 % dose reduction were noted in the maximum dose (D0.02 cc) of the left anterior descending artery, mean 12.3 Gy vs 18.1 Gy, and mean left ventricle V5Gy 13.2 % vs 41.1 %, respectively. The mean treatment duration was 451.5 sec with a median 8 breath-holds; 3 % junction locations between successive breath-holds were replicated. No locoregional or distant recurrences were observed in the 9-month median follow-up. Conclusion: Our workflow for DIBH with Helical-Tomotherapy addresses patient safety, treatment precision and challenges specific to this treatment unit. The workflow prevents junction issues by varying daily breath-hold durations and avoiding junction locations, providing a practical solution for left-sided breast cancer treatment with HT.