1. Executive Summary

Medication dispensing errors remain one of the most persistent patient safety challenges in hospital pharmacies worldwide. Despite rigorous training and double-check protocols, human verification at dispensing speeds of 3 to 5 seconds per pack carries an inherent error rate that no institution can fully eliminate through manual processes alone.

JKBEAT has developed an AI-powered drug verification system that addresses this gap at its source. The system employs a proprietary tri-modal fusion architecture combining AI visual recognition, laser inspection, and barcode scanning to verify each dispensed medication pack in 0.7 seconds with recognition accuracy of 99.99% or higher and zero missed detections.

This white paper presents the technical principles behind the tri-modal fusion approach, analyzes its performance characteristics in clinical deployment, and discusses its integration into hospital pharmacy workflows. The technology is currently operational at Grade 3A hospitals in China, including the First Affiliated Hospital of China Medical University in Shenyang.

The convergence of global regulatory trends — from China's 2026 mandatory drug traceability code implementation to the EU Falsified Medicines Directive and the US Drug Supply Chain Security Act (DSCSA) — makes automated verification an operational necessity rather than an optional enhancement.

2. The Patient Safety Challenge: Dispensing Errors in Hospital Pharmacies

2.1 The Scale of Dispensing Errors

Medication errors in hospital settings represent a significant burden on patient outcomes and healthcare costs. The World Health Organization's Global Patient Safety Challenge on Medication Without Harm has identified medication errors as one of the three priority areas for global patient safety improvement. Studies consistently report dispensing error rates ranging from 0.02% to 3.0% across different healthcare systems.

Manual verification, typically performed by a second pharmacist visually confirming each pack against the prescription order, operates at approximately 3 to 5 seconds per pack. Under workload pressure — particularly during morning dispensing peaks when inpatient medication carts must be prepared within tight time windows — verification thoroughness degrades. Research in human factors engineering has demonstrated that sustained visual inspection tasks suffer from vigilance decrements: detection performance declines significantly after approximately 20 to 30 minutes of continuous inspection.

Several factors compound the challenge:

• Visual similarity of drug packaging: Many medications share similar box dimensions, color schemes, and typography. A pharmacist checking hundreds of packs per hour faces significant visual fatigue.
• Complexity of multi-drug regimens: Inpatient prescriptions frequently involve 5 to 15 oral medications per patient. Each pack must be verified for drug identity, correct specification, and correct patient assignment.
• Traceability compliance overhead: The growing requirement to scan and upload traceability codes adds a procedural step to an already time-constrained workflow.

2.2 The Regulatory Push for Traceability

Governments worldwide are implementing mandatory drug traceability systems that require verification at the point of dispensing:

China: Starting from 2026, all drug distribution processes nationwide are required to collect and upload traceability codes under the principle of "scan every code, upload every scan." Hospital pharmacies must complete code scanning, verification, and uploading at the dispensing stage.

European Union: The Falsified Medicines Directive (FMD 2011/62/EU) requires serialization and verification at the point of dispensing across EU member states.

United States: The Drug Supply Chain Security Act (DSCSA) mandates full electronic, interoperable product tracing at the package level.

Southeast Asia and Middle East: Regulatory frameworks in Thailand, Saudi Arabia, and the UAE are progressively adopting traceability requirements modeled on EU and US precedents.

3. Technology Overview: Tri-Modal Fusion

3.1 Design Philosophy

The JKBEAT verification system is built on the principle that no single inspection modality can reliably detect all categories of dispensing errors. The tri-modal fusion architecture employs three independent inspection channels — AI visual recognition, laser inspection, and barcode scanning — and fuses their outputs in real time to produce a unified verification decision for each drug pack.

3.2 AI Visual Recognition

The visual recognition subsystem is the primary identification channel. It analyzes the external appearance of each drug pack using JKBEAT's fully self-developed AI vision algorithms, which are not derived from third-party modules.

The visual analysis extracts four categories of features from the drug packaging:

• Text content: Drug name, specification, manufacturer, lot number, and expiration date via OCR tuned for pharmaceutical packaging typography.
• Graphic elements: Logos, regulatory marks, and manufacturer trademarks provide additional identification signals.
• Color distribution: The overall color profile serves as a rapid pre-filtering signal, as drug manufacturers frequently use distinct color schemes.
• Shape and dimensions: The physical geometry of the pack provides a fourth verification dimension, effective for distinguishing between dosage forms and packaging sizes.

The AI vision algorithms are protected by five computer software copyrights registered with the National Copyright Administration of China, covering visual recognition of object surface defects (Registration No. 2022SR0312744), color analysis (Registration No. 2022SR0308960), and size measurement (Registration No. 2022SR1268016), along with the deblistering machine control software (Registration No. 2022SR1268059) and the verification machine intelligent control software (Registration No. 2022SR1268416).

3.3 Laser Inspection

The laser inspection subsystem operates in parallel with visual recognition, providing an independent physical verification channel:

• Thickness measurement: Each drug pack passes through a laser measurement stage that determines its physical thickness with high precision. A thickness deviation beyond tolerance triggers a rejection.
• Integrity verification: The laser system detects physical anomalies such as torn blister foils, crushed packaging, or abnormal bulging.

3.4 Barcode Scanning

The barcode scanning subsystem reads standard drug traceability barcodes (including GS1-128, GS1 DataMatrix, and Chinese domestic traceability code formats):

• Traceability code capture: The barcode is decoded and its traceability identifier is extracted for upload to the hospital's information system and national traceability platform.
• Identity cross-validation: The decoded barcode data is cross-referenced against the visual recognition result and the prescription order.

3.5 Real-Time Data Fusion Architecture

The three data streams are fused in real time through a decision engine that evaluates all available evidence before rendering a verification verdict:

1. Concordance requirement: For a pack to pass verification, the identity indicated by visual recognition must be consistent with the identity encoded in the barcode. The laser thickness measurement must fall within the expected range.
2. Multi-factor confidence scoring: Each channel produces a confidence score. The fusion engine computes a composite score, providing graceful degradation if one channel produces low confidence.
3. Pass/fail with audit trail: Every verification event is logged for audit purposes, including raw data from each channel, the fusion decision, and the timestamp.
4. Prescription order matching: The fused identity result is compared against the electronic prescription order to confirm clinically meaningful verification.

This architecture completes the full verification cycle — visual capture, laser measurement, barcode decode, data fusion, and order comparison — in 0.7 seconds per pack. The core mechanical and algorithmic design is protected by granted invention patent CN117963604B (ZL202410362593.X).

4. Performance Analysis

4.1 Verification Speed

The system processes each drug pack in 0.7 seconds, compared to 3 to 5 seconds for manual checking. This represents a 4 to 7 times improvement in per-pack throughput.

For a hospital pharmacy preparing 500 inpatient medication packs during a morning dispensing cycle, the verification time is reduced from approximately 25–42 minutes (manual) to under 6 minutes (automated).

4.2 Recognition Accuracy

The system achieves recognition accuracy of 99.99% or higher with zero missed detections:

• Recognition accuracy (≥ 99.99%): Measures the proportion of correctly identified drug packs among all packs processed.
• Zero missed detections: The system does not fail to inspect any drug pack that passes through the verification stage. Every pack is captured and processed by all three modalities.

4.3 Self-Developed AI Algorithms

The AI vision algorithms are fully developed in-house by JKBEAT's engineering team, providing several technical advantages:

• Domain-specific optimization: Trained specifically on pharmaceutical packaging recognition tasks rather than general-purpose vision modules.
• Continuous improvement: JKBEAT can iterate on model improvements without dependency on third-party technology providers.
• Intellectual property control: Protected by five registered computer software copyrights and one invention patent.
• Adaptability: Can be tailored to recognize drug packaging from different national markets, essential for international deployment.

5. Clinical Integration

5.1 Workflow Integration with Hospital Pharmacy Systems

The verification machine integrates into existing hospital pharmacy workflows as a standalone verification station. The typical integration architecture:

1. Prescription data receipt: The machine receives electronic prescription data from the hospital's pharmacy information system (PIS) via standard network interfaces.
2. Pack loading: Medication packs are presented to the verification chamber by a pharmacist or by integration with an upstream automated system.
3. Tri-modal verification: Each pack is processed through all three channels. The fused result is compared against the prescription order.
4. Pass/fail routing: Verified packs are sorted as pass or fail. Failed packs are segregated for pharmacist review with a detailed discrepancy report.
5. Data logging and upload: All verification events, including traceability barcode data, are logged for audit and regulatory compliance.

5.2 Traceability Compliance

The barcode scanning capability supports China's National Drug Traceability System requirements. For hospitals subject to the 2026 mandate, the verification machine serves a dual function: clinical verification and regulatory traceability — both in a single 0.7-second pass.

For international deployments, support for GS1-128 and GS1 DataMatrix formats provides compatibility with EU FMD and US DSCSA serialization frameworks.

5.3 Case Study: First Affiliated Hospital of China Medical University

The JKBEAT AI drug verification machine has been deployed at the First Affiliated Hospital of China Medical University in Shenyang, a Grade 3A (the highest tier) hospital.

Reported outcomes:

• Improved drug verification speed from 3–5 seconds per pack to 0.7 seconds per pack
• Reduced dispensing error rates compared to the manual verification baseline
• Enhanced medication safety through consistent, fatigue-free automated verification
• Full-process traceability with verification events logged and available for audit

JKBEAT's broader hospital footprint extends to additional Grade 3A institutions, including the Luzhou Maternal and Child Health Hospital (December 2024), the Shanghai Mental Health Center (November 2025), and the Zhanjiang No. 2 Traditional Chinese Medicine Hospital (July 2025), through authorized distributors such as Sinopharm Sichuan Medical Equipment Co., Ltd. and Guangdong Pharmaceutical Import & Export Co., Ltd.

6. Complementary Technology: Automatic Deblistering

6.1 The End-to-End Automation Concept

Drug verification in hospital pharmacies is typically preceded by a drug preparation step: removing individual tablets or capsules from blister packs (deblistering). JKBEAT offers the Automatic Deblistering Machine as a complementary product. Together, the two devices form an end-to-end automated pharmacy line:

• Deblistering: Removes tablets and capsules from blister packs at high speed
• Packaging: Tablets are packaged into patient-specific oral medication packs
• Verification: Each completed pack passes through the JKBEAT AI verification machine

6.2 Technical Advantages of Single-Source Integration

• Data continuity: Prescription data flows seamlessly from packaging to verification without format conversion.
• Synchronized operation: Deblistering and verification machines can be speed-matched for optimal throughput.
• Unified support: A single vendor provides maintenance, software updates, and technical support.

7. Global Applicability

7.1 Regulatory Landscape Across Regions

Region	Regulation	Verification Requirement
China	National Drug Traceability System (2026)	Per-pack traceability code scanning and upload at dispensing
European Union	Falsified Medicines Directive (FMD)	Serialization verification at point of dispensing
United States	Drug Supply Chain Security Act (DSCSA)	Electronic product tracing at package level
Thailand	Thai FDA drug traceability initiative	Progressive serialization and verification requirements
Saudi Arabia / UAE	GCC pharmaceutical regulations	Serialization and traceability aligned with international standards

7.2 Certification Roadmap

JKBEAT currently holds ISO 9001:2015 Quality Management System certification (Certificate No. CZ2511009QR1), covering the assembly and production of drug packaging machines, verification machines, and unpacking machines. The certification was first obtained on November 19, 2022, with the current certificate valid until November 16, 2028.

For international market access, JKBEAT's certification roadmap includes:

• CE marking: Conformity assessment for the European Economic Area
• SFDA/Thai FDA registration: Regulatory clearance for Southeast Asian markets
• Continued ISO 9001 maintenance: Annual surveillance audits and triennial recertification

8. Conclusion: The Future of Pharmacy Verification

The convergence of three trends — rising patient safety standards, expanding drug traceability regulations, and the maturation of AI-based visual recognition technology — is transforming pharmacy verification from a manual, error-prone process into an automated, data-driven quality control function.

The JKBEAT tri-modal fusion system represents a technically grounded approach to this transformation. By combining AI visual recognition, laser inspection, and barcode scanning in a single verification pass, the system achieves recognition accuracy of 99.99% or higher with zero missed detections at a throughput rate of 0.7 seconds per pack — a 4 to 7 times improvement over manual verification.

As global traceability regulations move from phased implementation to full enforcement, the operational question for hospital pharmacies is shifting from whether to automate verification to how to implement it most effectively.

---

About JKBEAT

JKBEAT (Liaoning Precision Control Beat Robot Co., Ltd.) was founded on December 23, 2021, in Shenyang, Liaoning Province, China. The company is focused exclusively on two vertical scenarios in hospital pharmacy automation: AI-powered drug verification and automatic deblistering. JKBEAT's intellectual property portfolio includes one granted invention patent (CN117963604B), five computer software copyrights, and ISO 9001:2015 quality management system certification.

Contact: www.jkbeat.com

---

References

All data points cited with [IA-XXXX-NNN] notation reference the JKBEAT Information Anchors Standard Manual v1.0 (2026-04-28), the company's single source of truth for externally communicated technical and commercial data.

This white paper is provided for informational purposes only. Product specifications and performance data are based on internal testing and clinical deployment results as documented in the JKBEAT Information Anchors Standard Manual. Actual performance may vary depending on deployment configuration and environmental conditions.