arXiv subjects

Xiangkun Xu

Publications and source records attributed to Xiangkun Xu.

12 recordsLinked to original sources

arXiv2026

DoseBridge: Denoising Diffusion Bridge Model for Dose Prediction in Lung Intensity-Modulated Proton Therapy

Most radiotherapy dose-prediction models use only CT images and anatomical structures, although intensity-modulated proton therapy (IMPT) dose also depends strongly on beam geometry and available clinical datasets are often small. We present DoseBridge, a denoising diffusion bridge model that uses the patient CT as a structured bridge endpoint and encodes plan-specific beam geometry in a spatially aligned beam mask. Multiscale fusion combines CT, target, organ-at-risk, and beam-mask representations with 1.95% additional parameters. DoseBridge was retrospectively evaluated on single-institution CT images and treatment plans from 52 patients with advanced-stage lung cancer treated with 60 Gy in 30 fractions; 42 cases were used for training and 10 for testing. Performance was assessed using image-similarity, dose-volume, and Lyman-Kutcher-Burman normal-tissue complication probability (NTCP) metrics and compared with two deep-learning models. On the test cohort, DoseBridge achieved a mean absolute error of 4.170 Gy, peak signal-to-noise ratio of 23.06 dB, and structural similarity index of 0.798, outperforming both comparison models on these metrics. Clinical target volume D95 differed from the reference dose by 0.62 +/- 1.6 Gy; signed organ-at-risk mean-dose differences ranged from -0.32 to 0.24 Gy, and NTCP differences were -0.40 +/- 2.2 and 0.52 +/- 3.4 percentage points for acute esophagitis and radiation pneumonitis, respectively. Changing only the beam mask redirected predicted low-dose entrance regions while preserving the high-dose target region. To our knowledge, DoseBridge is the first denoising diffusion bridge model for radiotherapy dose prediction. These results support its feasibility as a beam-aware planning prior for lung IMPT, pending evaluation in larger external cohorts.

cs.CV↗

arXiv2024

P/D-Serve: Serving Disaggregated Large Language Model at Scale

Serving disaggregated large language models (LLMs) over tens of thousands of xPU devices (GPUs or NPUs) with reliable performance faces multiple challenges. 1) Ignoring the diversity (various prefixes and tidal requests), treating all the prompts in a mixed pool is inadequate. To facilitate the similarity per scenario and minimize the inner mismatch on P/D (prefill and decoding) processing, fine-grained organization is required, dynamically adjusting P/D ratios for better performance. 2) Due to inaccurate estimation on workload (queue status or maintained connections), the global scheduler easily incurs unnecessary timeouts in prefill. 3) Block-fixed device-to-device (D2D) KVCache transfer over cluster-level RDMA (remote direct memory access) fails to achieve desired D2D utilization as expected. To overcome previous problems, this paper proposes an end-to-end system P/D-Serve, complying with the paradigm of MLOps (machine learning operations), which models end-to-end (E2E) P/D performance and enables: 1) fine-grained P/D organization, mapping the service with RoCE (RDMA over converged ethernet) as needed, to facilitate similar processing and dynamic adjustments on P/D ratios; 2) on-demand forwarding upon rejections for idle prefill, decoupling the scheduler from regular inaccurate reports and local queues, to avoid timeouts in prefill; and 3) efficient KVCache transfer via optimized D2D access. P/D-Serve is implemented upon Ascend and MindSpore, has been deployed over tens of thousands of NPUs for more than eight months in commercial use, and further achieves 60\%, 42\% and 46\% improvements on E2E throughput, time-to-first-token (TTFT) SLO (service level objective) and D2D transfer time. As the E2E system with optimizations, P/D-Serve achieves 6.7x increase on throughput, compared with aggregated LLMs.

cs.DC↗

arXiv2024

A universal bioluminescence tomography system for pre-clinical image-guided radiotherapy research

CBCT-guided small animal irradiators encounter challenges in localizing soft-tissue targets due to low imaging contrast. Bioluminescence tomography (BLT) offers a promising solution, but they have largely remained in laboratorial development, limiting accessibility for researchers. In this work, we develop a universal, commercial-graded BLT-guided system (MuriGlo) designed to seamlessly integrate with commercial irradiators and empower researchers for translational studies. We demonstrate its capabilities in supporting in vitro and in vivo studies. The MuriGlo comprises detachable mouse bed, thermostatic control, mirrors, filters, and CCD, enabling multi-projection and multi-spectral imaging. We evaluate that the thermostatic control effectively sustains animal temperature at 37{\deg}C throughout imaging, and quantify that the system can detect as few as 61 GL261-AkaLuc cells in vitro. To illustrate how the MuriGlo can be utilized for in vivo image-guided research, we present 3 strategies, BLT-guided 5-arc, 2-field box, and BLI-guided single-beam, ranging from complicated high-conformal to simplest high-throughput plans. The high conformal BLT-guided 5-arc plan fully covers the gross tumor volume (GTV) at prescribed dose with minimal normal tissue exposure (3.9%), while the simplified, high-throughput BLT-guided 2-field box achieves 100% GTV coverage but results in higher normal tissue exposure (13.1%). Moreover, we demonstrate that the localization accuracy of MuriGlo for both widely-used SARRP and SmART irradiators is within1 mm, and the tumor coverage reaches over 97% with 0.75mm margin. The universal BLT-guided system offers seamless integration with commercial irradiators, achieving comparable localization accuracy, expected to supporting high-precision radiation research.

physics.med-ph↗

arXiv2020

Quantitative Bioluminescence Tomography-guided System for Conformal Irradiation In Vivo

Although cone-beam CT (CBCT) has been used to guide irradiation for pre-clinical radiotherapy(RT) research, it is limited to localize soft tissue target especially in a low imaging contrast environment. Knowledge of target shape is a fundamental need for RT. Without such information to guide radiation, normal tissue can be irradiated unnecessarily, leading to experimental uncertainties. Recognition of this need led us to develop quantitative bioluminescence tomography (QBLT), which provides strong imaging contrast to localize optical targets. We demonstrated its capability of guiding conformal RT using an orthotopic bioluminescent glioblastoma (GBM) model. With multi-projection and multi-spectral bioluminescence imaging and a novel spectral derivative method, our QBLT system is able to reconstruct GBM with localization accuracy <1mm. An optimal threshold was determined to delineate QBLT reconstructed gross target volume (GTV_{QBLT}), which provides the best overlap between the GTV_{QBLT} and CBCT contrast labeled GBM (GTV), used as the ground truth for the GBM volume. To account for the uncertainty of QBLT in target localization and volume delineation, we also innovated a margin design; a 0.5mm margin was determined and added to GTV_{QBLT} to form a planning target volume (PTV_{QBLT}), which largely improved tumor coverage from 75% (0mm margin) to 98% and the corresponding variation (n=10) of the tumor coverage was significantly reduced. Moreover, with prescribed dose 5Gy covering 95% of PTV_{QBLT}, QBLT-guided 7-field conformal RT can irradiate 99.4 \pm 1.0% of GTV vs. 65.5 \pm 18.5% with conventional single field irradiation (n=10). Our QBLT-guided system provides a unique opportunity for researchers to guide irradiation for soft tissue targets and increase rigorous and reproducibility of scientific discovery.

physics.med-ph↗

arXiv2019

Breaking the quantum adiabatic speed limit by jumping along geodesics

Quantum adiabatic evolutions find a broad range of applications in quantum physics and quantum technologies. The traditional form of the quantum adiabatic theorem limits the speed of adiabatic evolution by the minimum energy gaps of the system Hamiltonian. Here, we experimentally show using a nitrogen-vacancy center in diamond that, even in the presence of vanishing energy gaps, quantum adiabatic evolution is possible. This verifies a recently derived necessary and sufficient quantum adiabatic theorem and offers paths to overcome the conventionally assumed constraints on adiabatic methods. By fast modulation of dynamic phases, we demonstrate near-unit-fidelity quantum adiabatic processes in finite times. These results challenge traditional views and provide deeper understanding on quantum adiabatic processes, as well as promising strategies for the control of quantum systems.

quant-ph↗

arXiv2018

Observation of Non-Markovianity at Room Temperature by Prolonging Entanglement in Solids

The non-Markovia dynamics of quantum evolution plays an important role in open quantum sytem. However, how to quantify non-Markovian behavior and what can be obtained from non- Markovianity are still open questions, especially in complex solid systems. Here we address the problem of quantifying non-Markovianity with entanglement in a genuine noisy solid state system at room temperature. We observed the non-Markovianity of quantum evolution with entanglement. By prolonging entanglement with dynamical decoupling, we can reveal the non-Markovianity usually concealed in the environment and obtain detailed environment information. This method is expected to be useful in quantum metrology and quantum information science.

quant-ph↗

arXiv2016

Experimental Adiabatic Quantum Factorization under Ambient Conditions Based on a Solid-State Single Spin System

The adiabatic quantum computation is a universal and robust method of quantum computing. In this architecture, the problem can be solved by adiabatically evolving the quantum processor from the ground state of a simple initial Hamiltonian to that of a final one, which encodes the solution of the problem. By far, there is no experimental realization of adiabatic quantum computation on a single solid spin system under ambient conditions, which has been proved to be a compatible candidate for scalable quantum computation. In this letter, we report on the first experimental realization of an adiabatic quantum algorithm on a single solid spin system under ambient conditions. All elements of adiabatic quantum computation, including initial state preparation, adiabatic evolution, and final state readout, are realized experimentally. As an example, we factored 35 into its prime factors 5 and 7 on our adiabatic quantum processor.

quant-ph↗

arXiv2015

Resolving Remoter Nuclear Spins in a Noisy Bath by Dynamical Decoupling Design

We experimentally resolve several weakly coupled nuclear spins in diamond using a series of novelly designed dynamical decoupling controls. Some nuclear spin signals, hidden by decoherence under ordinary dynamical decoupling controls, are shifted forward in time domain to the coherence time range and thus rescued from the fate of being submerged by the noisy spin bath. In this way, more and remoter single nuclear spins are resolved. Additionally, the field of detection can be continuously tuned on sub-nanoscale. This method extends the capacity of nanoscale magnetometry and may be applicable in other systems for high-resolution noise spectroscopy.

quant-ph↗

arXiv2013

NV-Center Based Digital Quantum Simulation of a Quantum Phase Transition in Topological Insulators

Nitrogen-vacancy centers in diamond are ideal platforms for quantum simulation, which allows one to handle problems that are intractable theoretically or experimentally. Here we propose a digital quantum simulation scheme to simulate the quantum phase transition occurring in an ultrathin topological insulator film placed in a parallel magnetic field [Zyuzin \textit{et al.}, Phys. Rev. B \textbf{83}, 245428 (2011)]. The quantum simulator employs high quality spin qubits achievable in nitrogen-vacancy centers and can be realized with existing technology. The problem can be mapped onto the Hamiltonian of two entangled qubits represented by the electron and nuclear spins. The simulation uses the Trotter algorithm, with an operation time of the order of 100 $μ$s for each individual run.

quant-ph↗

arXiv2013

Observation of Time-domain Rabi Oscillations in the Landau-Zener Regime with a Single Electronic Spin

Under resonant conditions, a long sequence of landau-zener transitions can lead to Rabi oscillations. Using a nitrogen-vacancy (NV) center spin in diamond, we investigated the interference between more than 100 Landau-Zener processes. We observed the new type of Rabi oscillations of the electron spin resulting from the interference between successive Landau-Zener processes in various regimes, including both slow and fast passages. The combination of the control techniques and the favorable coherent properties of NV centers provides an excellent experimental platform to study a variety of quantum dynamical phenomena.

quant-ph↗

arXiv2012

Coherence-protected Quantum Gate by Continuous Dynamical Decoupling in Diamond

To implement reliable quantum information processing, quantum gates have to be protected together with the qubits from decoherence. Here we demonstrate experimentally on nitrogen-vacancy system that by using continuous wave dynamical decoupling method, not only the coherence time is prolonged by about 20 times, but also the quantum gates is protected for the duration of controlling time. This protocol shares the merits of retaining the superiority of prolonging the coherence time and at the same time easily combining with quantum logic tasks. It is expected to be useful in task where duration of quantum controlling exceeds far beyond the dephasing time.

quant-ph↗

arXiv2011

Landau-Zener-Stuckelberg interferometry of a single electron spin in a noisy environment

We demonstrate quantum coherent control of a single electron spin in a NV center in diamond using the Landau-Zener-Stuckelberg interferometry at room temperature. Interference pattern is observed oscillating as a function of microwave frequency. The decays in the visibility of the interference are well explained by numerical simulation which includes the thermal fluctuations of the nuclear bath which shows that Landau-Zener-Stuckelberg interferometry can be used for probing electron spin decoherence processes.

quant-ph↗