DNA testing methods have become incredibly sensitive. Many court stakeholders have accordingly shifted their focus from questions about "who is the source of the DNA?" to "how did the DNA get to be where it was found?" Given this shift, some in the forensic DNA community have urged courts in the United States to permit testimony interpreting DNA evidence given activity-level propositions (ALP). These advocates assert that ALP reporting and testimony will "help" courts answer the relevant questions in their cases. This assertion requires scrutiny. Examinations of wrongful convictions teach us that forensic experts can do irreparable harm when attempting to help courts answer difficult questions. The potential that ALP reporting and testimony will contribute to miscarriages of justice in U.S. courts must be considered. The commentary in this article critically assesses the claim that ALP reporting and testimony will help courts in the United States reach the correct outcome. We do so by analyzing an activity-level "lookup table" that was recently published by advocates of ALP reporting in a paper titled Interpreting DNA under fingernails given activity level propositions. Our analysis considers the potential that LRs reported in the lookup table will mislead end-users. We also consider the vulnerability of U.S. courts to wrongful convictions based on overstated forensic science testimony. Our analysis ultimately reveals that application of the lookup table to casework in the United States may not help end-users reach the correct outcome, and may, in fact, increase the risk of wrongful convictions.
Video dehazing aims to restore clean scenarios from a sequence of hazy frames, where frame alignment is a critical stage for leveraging temporal information. However, haze degrades contrast and obscures details, making alignment challenging. Existing methods ignore the impairment of haze on alignment and thus struggle to align frames accurately. To address this challenge, we propose an alignment network with the temporal lookup table (temporal-LUT), which effectively enhances the haze-degraded frames and provides vivid cues for precise alignment. Specifically, to tackle the color degradation of haze, we employ a learnable lookup table (LUT) to enhance hazy color. The color mapping nature of LUT favorably preserves the naturalness of enhanced outcomes. Besides, we introduce a temporal weight prediction strategy to strengthen inter-frame interaction, which ensures temporal consistency across enhanced results and thereby benefits alignment. Extensive experimental results on two widely used benchmarks and real-world scenes demonstrate the superiority of our method.
Distortion product otoacoustic emissions (DPOAEs) are weak cochlear tones (f 3 = 2f 1 - f 2) evoked by two primary tones (f 1, f 2) and used clinically to assess cochlear function. In practice, f 1 and f 2 are typically presented using two independent sound sources. This is because using one transducer to output both tones can generate intermodulation distortion (IMD) at the same frequency (f 3) as the DPOAE due to transducer nonlinearity, which makes it difficult to distinguish measurement artifacts from biological emissions. As a result, DPOAE measurement is difficult with conventional earphones, which typically use one playback transducer per ear. Here, we describe a method that enables single-transducer DPOAE measurements by adding a cancellation tone at f 3 and tuning its amplitude and phase (via a lock-in amplifier technique) to reduce IMD. Since IMD is level dependent and in-ear stimulus levels vary across ears (due to ear-canal acoustics), the required cancellation parameters are therefore stored in a level-specific lookup table (LUT). The proposed method:•Reduces IMD at the DPOAE frequency to enable single-transducer measurement.•Uses LUT parameters to handle level-dependent IMD.•Was validated on 40 ears against a two-transducer baseline.
Measuring the optical properties of small-volume turbid samples remains a significant challenge. This work presents the experimental optimization and validation of a frequency-domain diffuse optical method applied to media in a standard cuvette geometry. We assessed the impact of boundary conditions, determining that reflective cuvette walls provide superior precision and reproducibility compared to absorbing walls. Additionally, we developed a physics-informed lookup table based on diffusion theory and calibrated it using experimental data from optically characterized samples. Results demonstrate a precision of about 0.3% for the measured optical data and accuracies of approximately 20% for absorption and 10% for scattering using only four calibration samples, confirming the method's feasibility for milliliter-scale optical characterization. To our knowledge, this represents the first experimental validation of a cuvette-based system for the quantitative optical characterization of turbid media.
Representing a set of k-mers-strings of length k-in small space under fast lookup queries is a fundamental requirement for several applications in Bioinformatics. A data structure based on sparse and skew hashing (SSHash) was recently proposed for this purpose (Pibiri 2022): it combines good space effectiveness with fast lookup and streaming queries. It is also order-preserving, i.e. consecutive k-mers (sharing a prefix-suffix overlap of length k-1) are assigned consecutive hash codes which helps compressing satellite data typically associated with k-mers, like abundances and color sets in colored De Bruijn graphs. We study the problem of accelerating queries under the sparse and skew hashing indexing paradigm, without compromising its space effectiveness. We propose a refined data structure with less complex lookups and fewer cache misses. We give a simpler and faster algorithm for streaming lookup queries. The refined architecture translates to substantial performance gains, outperforming the original version of SSHash in both index construction speed and query efficiency. Compared to indexes with similar capabilities and based on the Burrows-Wheeler transform, like SBWT and FMSI, SSHash is significantly faster to build and query. SSHash is competitive in space with the fast (and default) modality of SBWT when both k-mer strands are indexed. While larger than FMSI, it is also more than one order of magnitude faster to query. The SSHash software is available at https://github.com/jermp/sshash, and also distributed via Bioconda. A benchmark of data structures for k-mer sets is available at https://github.com/jermp/kmer_sets_benchmark. The datasets used in this article are described and available at https://zenodo.org/records/17582116.
Assigning area-level socioeconomic measures to street addresses across historical and contemporary census years requires harmonising address, geography and deprivation datasets that change over time. This methods article describes a reproducible workflow and online implementation for assigning New Zealand Index of Socioeconomic Deprivation (NZDep) measures to street addresses for census years 1991-2023. The method integrates public Land Information New Zealand address data, openly licensed postcode boundary polygons, Stats NZ census geographies and University of Otago NZDep datasets into an indexed SQLite lookup database and Shiny address-lookup tool (https://bit.ly/nzdep). Address labels are standardised, address points are spatially linked to year-specific census geographies, and geography identifiers are joined to corresponding NZDep values. Raw NZDep scores are converted to census-year-specific national percentile ranks, allowing relative deprivation position to be expressed on a consistent 1-100 scale across census periods. The workflow supports unit-level address lookup where available and base street-address lookup only where deprivation assignment is unambiguous. The final database contained 2405,589 address records and achieved 99.65% completeness across 16,839,123 potential address-year NZDep assignments.
Clinicians' adoption of interoperability tools influences care quality, but evidence of actual use is limited. We analyze clinicians' use of outside records delivered via Epic Care Everywhere (CE), focusing on use frequency and predictors such as gender, experience, specialty, and role. Differences between pre-pandemic (2018-2019) and pandemic (2020-2021) periods are also examined to see how COVID-19 affected use of outside records. De-identified EHR metadata from UCSF clinicians (n = 1442) during pre-pandemic and pandemic periods, totaling 686 797 clinician-day observations, were analyzed. We measured usage intensity (mean CE lookups per appointment) and breadth (percentage of appointments with ≥1 lookup). Generalized linear models (GLMs) with a negative binomial distribution for overdispersion were used to estimate predictors of intensity. CE usage intensity rose by 43.0% post-COVID-19 onset compared with the pre-pandemic. Clinician specialty most strongly predicted use, with Nephrology and Cardiology showing the highest breadth (56.0% and 53.0% of visits, respectively), while Dermatology (13.2%) and Pediatrics (23.4%) were lowest. Residents used CE at 23.0% greater intensity than attendings, and each additional year of experience was linked to a 0.57% decrease in intensity. Clinician use of interoperability tools was higher in specialties such as Nephrology and Cardiology that require more care coordination, and among less experienced clinicians including resident physicians. Use increased after the pandemic began, likely due to ongoing adoption trends and increased clinical demands during system strain and uncertainty. These findings underscore the critical importance of considering clinician behavior and contextual factors, such as specialty care needs, in addition to technical capabilities, when promoting the adoption and use of interoperability tools.
Medical-device imaging workflows need a reproducible way to connect unique device identification (UDI) with Digital Imaging and Communications in Medicine (DICOM) equipment metadata and with external evidence such as installation-acceptance or calibration records. The problem addressed here is not to build a hospital-wide platform. It is to define a minimal, checkable profile that preserves the full UDI, derives a device identifier suitable for registry lookup, and links the result to reviewable evidence. We derived a minimal UDI-DICOM mapping profile from International Medical Device Regulators Forum (IMDRF) UDI guidance, current DICOM equipment and UDI provisions, and U.S. Food and Drug Administration Global Unique Device Identification Database (GUDID) registry semantics. The revised contribution is not the individual standard fields themselves; it is the combination of a machine-readable manifest, explicit cross-layer validation rules, and a prototype validation artifact. The artifact accepts a manifest plus DICOM metadata, performs presence and parseability checking, registry-resolution checking, cross-layer consistency checking, and evidence-link closure checking, and emits reviewable Markdown, JSON, and JSON Lines outputs. Validation was executed on an offline fixture, a live openFDA/GUDID lookup path, and a synthetic pydicom-generated DICOM fixture. All available artifact runs passed. No real-device DICOM specimen has yet been validated. This paper therefore contributes a bounded Methods Paper artifact for imaging informatics, biomedical engineering traceability, and standards-oriented workflow integration readers, while explicitly excluding claims of clinical validation, regulatory approval, hospital-wide deployment, or universal interoperability.
BeanGPT is a domain-specific retrieval augmented generation system designed to support research and breeding decisions in common bean (Phaseolus vulgaris L.) by transforming natural language questions into citation-backed, verifiable answers. The platform integrates a large, curated corpus of legume-focused peer-reviewed literature with structured multi-year agronomic trial records collected across diverse environments, climate projections extending to 2090 under multiple emission scenarios, and standardized cultivar nomenclature to resolve naming inconsistencies across datasets and publications. BeanGPT combines semantic retrieval from a vector database with intent-based query routing and structured parameter extraction to direct questions to genetics, field performance analytics, or climate modules. To reduce errors that commonly occur in general-purpose language models, BeanGPT incorporates a genomic index that enables constant time membership lookup of gene and protein identifiers against authoritative resources, ensuring that molecular entities are either validated or clearly flagged as literature-derived. The system is implemented with a streaming web interface and an asynchronous backend that supports concurrent users and can generate interactive visualizations through automated Plotly code generation. Beta testing demonstrated strong retrieval relevance, low response latency, reliable gene verification, and high citation precision, indicating that domain-grounded RAG can improve accuracy and usability for Phaseolus vulgaris research.
Background/Objectives: Outcomes after CD19-directed chimeric antigen receptor (CAR) T-cell therapy for relapsed or refractory (R/R) aggressive large B-cell lymphoma (aLBCL) remain heterogeneous. Tumor genomic biomarkers, such as TP53 alteration, MYC/BCL2/BCL6 rearrangement-defined double-hit or triple-hit lymphoma (DHL/THL), cell of origin (COO), and complex karyotype, are established or candidate prognostic factors in conventionally treated lymphoma, but their relevance after CAR T-cell therapy is uncertain. We conducted a systematic review with exploratory meta-analysis of biomarker-stratified outcomes after CD19 CAR T-cell therapy in aLBCL. Methods: We searched MEDLINE, Embase, and Web of Science/BIOSIS (April 2026), with targeted PubMed citation lookup during full-text retrieval (PROSPERO CRD420261350514). Eligible studies enrolled adults with R/R disease treated with protocol-eligible CD19 CAR T-cell therapy and reported prespecified tumor genomic biomarkers with stratified outcomes. Random-effects models, using restricted maximum-likelihood estimation with Hartung-Knapp-Sidik-Jonkman (HKSJ) adjustment, were fitted when at least three comparable, non-overlapping studies provided extractable data. Results: After duplicate removal, 182 records were screened, 37 were assessed for eligibility, and 26 studies were included in the qualitative synthesis; 10 contributed to 4 pooled analyses. DHL/THL-positive disease was associated with worse unadjusted overall survival (OS) (hazard ratio [HR] 1.52; 95% confidence interval [CI], 1.21-1.89; 95% prediction interval (PI), 0.56-4.08), and non-Germinal center B-cell-like (GCB)/ABC COO with worse adjusted progression-free survival (PFS) (HR 1.44; 95% CI, 1.04-2.00; 95% PI, 0.86-2.43). The complete-response analyses for TP53 alteration (OR 1.30; 95% CI, 0.01-156.60) and COO (OR 1.27; 95% CI, 0.24-6.61) were statistically uninformative. No study permitted evaluation of complex karyotypes. Conclusions: Biomarker-stratified evidence after CD19 CAR T-cell therapy is sparse and inconsistently reported. DHL/THL status and non-GCB/activated B-cell-like (ABC) COO showed exploratory survival signals, whereas the TP53 and COO complete-response analyses were uninformative. These biomarkers remain hypothesis-generating rather than validated predictors of CAR T-cell outcome, and standardized, prospective biomarker-stratified reporting is needed.
While coding regions in the genome have a direct interpretation in terms of protein products, significant fractions are non-coding and yet control essential biological functions. Unlike the genetic code, there is no "lookup table" that identifies where regulatory proteins, known as transcription factors (TFs), bind. Here, we extract these binding sites by distilling sequences of nucleotide letters into collective coordinates (hyperletters) representing the binding sites that are active under specific environmental conditions. Going beyond local information footprints between individual bases and expression levels, our information blueprint algorithm compresses the global information by optimising filters that simultaneously scan an entire promoter sequence. Inspired by renormalisation-group techniques, we identify TF binding sites as coarse-grained variables combining groups of correlated mutations with the highest collective impact on gene expression. We validate our approach on experimental data for E. coli and discover novel regulatory elements illustrating its deployment at scale across growth conditions.
This paper proposes a probabilistic shaping modulation scheme based on index-driven constellation partitioning. The 32QAM constellation is divided into eight non-overlapping energy-level patterns, with every three consecutive subcarriers grouped as a mapping unit controlled jointly by index bits through a predefined lookup table. By assigning more subcarriers to low-energy constellation modes located in the inner circle, the activation probability of low-energy symbols is increased, thereby reducing the average transmit power. Experimental validation is conducted over a 2 km seven-core fiber optic system. Experimental results demonstrate that at a bit error rate (BER) of 3.8 × 10-3, the proposed index-driven constellation-shaped signal achieves up to 0.53 dB gain compared to uniform 32QAM signals. Meanwhile, the BER performance variation among different modes remains below 0.5 dB. With low implementation complexity, the proposed scheme exhibits promising potential in multi-user optical access scenarios.
Objectives: When a Candida species is identified in an ICU patient, susceptibility results are typically available in 24-72 h. In this study, we built a machine learning model using four variables available at identification to estimate resistance probability in real time. Methods: We analysed 747 fungal isolates from 725 ICU patients (January 2021-March 2026). We trained and compared a Random Forest and a Logistic Regression model, evaluating both with temporal cross-validation, permutation feature importance, three-category (S/I/R) prediction, and calibration analysis. Results: Multidrug resistance doubled from 24.5% (2021) to 51.1% (2025), and Candida auris grew eight-fold in three years. Random Forest reached AUC 0.885 on the held-out test set and 0.848 on prospective 2024-2025 data (Brier score 0.093). Species identity and drug choice together explained 87% of predictive signal. Local C. albicans fluconazole resistance (~16%) far exceeded the ECMM European figure of 0%, and C. krusei was four times more prevalent than the continental average. Conclusions: A four-variable model may provide calibrated resistance estimates during the critical gap before susceptibility results return, though performance reflects predominantly deterministic species-drug patterns rather than complex learned biology. Overall performance was comparable to a rule-based lookup table, confirming that the majority of predictive signal derives from established species-drug susceptibility patterns. Meaningful added value is limited to temporal trend tracking and improved prediction where resistance is acquired rather than intrinsic (C. albicans, C. tropicalis hard-subset AUC 0.929 vs. rule-based 0.899). The model complements a local antifungal testing; it does not replace one.
Since self-attention layers in Transformers are permutation invariant by design, positional encodings must be explicitly incorporated to enable spatial understanding. However, fixed-size lookup tables used in traditional learnable position embeddings (PEs) limit extrapolation capabilities beyond pre-trained sequence lengths. Expert-designed methods such as ALIBI and ROPE, mitigate this limitation but demand extensive modifications for adapting to new modalities, underscoring fundamental challenges in adaptability and scalability. In this work, we present SEQPE, a unified and fully learnable position encoding framework that represents each n-dimensional position index as a symbolic sequence and employs a lightweight sequential position encoder to learn their embeddings in an end-to-end manner. To regularize SEQPE's embedding space, we introduce two complementary objectives: a contrastive objective that aligns embedding distances with a predefined position-distance function, and a knowledge distillation loss that anchors out-of-distribution position embeddings to in-distribution teacher representations, further enhancing extrapolation performance. Experiments across language modeling, long-context question answering, and 2D image classification demonstrate that SeqPE not only surpasses strong baselines in perplexity, exact match (EM), and accuracy-particularly under context length extrapolation-but also enables seamless generalization to multi-dimensional inputs without requiring manual architectural redesign.
Hospital information technology (IT) outages severely disrupt clinical workflows and use of electronic medical records, threatening patient safety and operational continuity. Traditional disaster response training faces limitations including high resource requirements, restricted repeatability, and inability to be conducted without interrupting 24/7 hospital operations. Digital twin technology enables realistic, repeatable simulation training in virtual environments, avoiding operational disruption. This study developed and implemented a digital twin-based virtual hospital platform for Level 1 IT outage disaster response training and evaluated its feasibility through quantitative performance metrics and participant survey feedback. A digital twin-based virtual hospital platform modeling 317 clinical spaces and 6 building entrances of Yongin Severance Hospital, South Korea, was developed to simulate a hospital information system failure (Code White Level 1 IT outage) with 7 patient cases of varying complexity levels, covering complete outpatient workflows from registration through payment. Sixty multidisciplinary participants (physicians, nurses, laboratory technicians, pharmacists, and administrative staff) were recruited through purposive sampling from clinical departments and support services directly involved in outpatient IT outage response. Emergency prescription and patient information lookup systems were integrated into the training. Performance evaluation included scenario completion rates, prescription accuracy, completion times, and operational readiness scores. Training outcomes were compared with 2023 conventional training records from the same institution using descriptive metrics. Open-ended survey responses were analyzed using structured content summarization with text mining and word cloud techniques. A 7-item operational readiness checklist was assessed by a panel of 5 training facilitators to evaluate system functional completeness. In July 2024, 60 multidisciplinary participants completed the training exercise. All 7 patient scenarios achieved 100% completion rates with perfect accuracy in medical billing concordance and prescription entry timeliness. Scenario completion times ranged from 25 to 47 minutes, with variations reflecting testing wait times and workflow complexity. The overall operational readiness score was 80% (with 70% for digital twin platform operational readiness and 90% for emergency prescription program operational readiness). Training reduced resource consumption by 70 minutes compared to the 2023 conventional training approach, decreasing full-time equivalent requirements from 0.072 to 0.038. Open-ended survey feedback yielded five content categories (target extensions, mock training ideas, drug/prescription system improvement, operations and evaluation systems, and process improvement), with realism, collaboration, and prescription as the most frequently cited keywords. This proof-of-concept study demonstrates that a simulation-oriented digital twin platform can support multi-department IT disaster response training with complete workflow execution. Identified technical gaps in user permissions and prescription classification provide a concrete development roadmap for institutional deployment.
Cervical cancer remains a major global health burden, particularly in underserved populations where late diagnoses contribute to high mortality rates. Accurate, early risk prediction is essential for improving outcomes and guiding preventive care. In this study, we introduce CERV-Score, a hybrid machine learning framework that advances prior approaches by combining structured clinical risk factors with recurrence-based genomic markers to generate continuous, probabilistic risk scores rather than traditional binary classifications. This enables nuanced patient stratification into low, moderate, and high-risk categories, providing clinicians with more actionable insights. Unlike previous models, CERV-Score integrates genomic recurrence analysis identifying genes consistently expressed across multiple RNA-seq samples to improve biological relevance and robustness. Additionally, we developed an interactive clinical-genomic decision support tool that delivers real-time, percentage-based risk predictions and includes a gene lookup function, bridging clinical practice and molecular exploration in a single platform. The hybrid CERV-Score model achieved high predictive performance (accuracy = 94.1%, F1 - score = 0.91, AUC = 0.94). Bootstrap resampling (1000 iterations) applied to the test predictions produced a 95% confidence interval for accuracy of 92.8%-95.4%, confirming the stability and robustness of the model's performance. These results highlight the contribution of probabilistic scoring, recurrence-driven genomic integration, and interactive visualization to enhance both accuracy and usability. By combining methodological innovation with practical clinical utility, CERV-Score represents a meaningful step beyond existing hybrid models, laying the groundwork for more interpretable, personalized, and deployable cervical cancer risk prediction systems.
Autonomous driving perception demands low latency, high temporal resolution, and stringent hardware efficiency. While event-based spiking neural networks (SNNs) offer bio-inspired sparse computation, their deployment on edge field-programmable gate arrays (FPGAs) is obstructed by irregular execution patterns and temporal state storage overhead. To address this, we propose HAPQ, a unified hardware-aware pruning and quantization pipeline for compact event-based object detection. Starting from an end-to-end adaptive sampling SNN detector (EAS-SNN), HAPQ conducts hardware-aware configuration search within discrete digital signal processor (DSP) and block RAM (BRAM) budgets, applies single-instruction-multiple-data (SIMD)-aligned structured pruning for computational regularity, and jointly quantizes synaptic weights and membrane potentials via a shift-friendly fixed-point recurrence. Evaluation on the Prophesee Gen1 dataset and an FPGA accelerator shows that HAPQ improves detection accuracy from 0.284 to 0.425 in mean average precision (mAP50:95) and achieves 0.722 mAP50. Hardware implementation reveals a reduction in lookup table (LUT) usage to 1680, complete DSP elimination, and a maximum operating frequency of 920.81 MHz at 0.630 W. These results confirm that effective temporal SNN deployment requires joint optimization of model architecture, state precision, and hardware-aligned workload organization.
Lateral stability of distributed drive electric vehicles (DDEVs) under high-speed and low-adhesion conditions is often evaluated using autonomous phase-plane analysis, which does not explicitly account for closed-loop yaw-moment control and thus yields conservative stability limits. This paper proposes a unified lateral-stability framework that integrates controllable-region (CR) analysis with torque-distribution mode selection under closed-loop control. CRs associated with no control, the differential braking distribution mode (DBDM), and the balanced torque-vectoring distribution mode (BTVDM) are constructed on the sideslip-angle-sideslip-angle-rate phase plane, while control efficiency is assessed in terms of convergence time and execution cost. A gradient-boosted decision-tree ensemble trained on nonlinear vehicle simulations is distilled into a lightweight four-dimensional lookup table for real-time implementation. Results show that, under a representative high-speed and low-adhesion condition (vx=105km/h, μ=0.3), the uncontrolled vehicle fails to satisfy tcmax=3s, whereas DBDM and BTVDM converge within 2.17 s and 2.32 s, respectively. DBDM provides faster recovery near high-risk boundary states, while BTVDM reduces the maximum longitudinal speed loss from 5.96 m/s to 1.40 m/s in closed-loop simulation. The proposed adaptive distribution mode reduces the speed loss by approximately 57.4% compared with pure DBDM, while maintaining comparable peak and RMS sideslip-angle and yaw-rate errors in both simulation and HiL tests.
We present an algorithm for efficient evaluation of Boys functions F0,…,Fkmax tailored to modern computing architectures, in particular graphical processing units, where maximum throughput is high and data movement is costly. The method combines rational minimax approximations with upward and downward recurrence relations. The non-negative real axis is partitioned into three regions, [0, ∞⟩ = A ∪ B ∪ C, where regions A and B are treated using rational minimax approximations and region C by an asymptotic approximation. This formulation avoids lookup tables and irregular memory access, making it well-suited for hardware with high maximum throughput and low latency. The rational minimax coefficients are generated using the rational Remez algorithm. For a target maximum absolute error of ɛtol = 5 × 10-14, the corresponding approximation regions and coefficients for Boys functions F0, …, F32 are provided in Appendix D.
Objective.This study aims to develop and validate a depth-of-interaction (DOI) positron emission tomography (PET) detector based on a bismuth germanate (BGO) crystal array for high-sensitivity small-animal PET imaging, and to investigate the feasibility of using a light-sharing window (LSW) structure for DOI encoding in BGO detectors despite their inherently low light yield.Approach.ABGO array with individual crystal dimensions ofmmwas coupled to a multi-pixel photon counter array using acrystal-to-sensor configuration. A 6 mm air-gap LSW was introduced between adjacent crystal pixels to generate DOI-dependent light sharing. DOI performance was evaluated using a mechanical collimation experiment, and a vertical line-source irradiation method combined with histogram matching was further implemented to generate Ratio-to-DOI lookup tables from a single static acquisition.Main results.The proposed detector achieved an intrinsic DOI resolution of 4.4 mm full width at half maximum (FWHM), a mean absolute error of 2.2 mm, and an average energy resolution of 17.4% at 511 keV. Using the vertical line-source irradiation calibration method, a DOI resolution of 4.9 mm FWHM was obtained in a single measurement, demonstrating comparable performance to the mechanical collimation calibration with substantially reduced calibration complexity.Significance.These results demonstrate the feasibility of applying the LSW method to low-light-yield BGO detectors and provide a practical, cost-effective DOI solution for high-sensitivity small-animal PET imaging.