# Voice-based versus RFID needle counting: how smart acupuncture counting architectures differ

> Dalin Tzu Chi Hospital and Chung Cheng University's voice-based counting system, Taipei Veterans General Hospital's wearable-sensor counting, and per-needle physical identification each trade off differently in record source and verification. This article analyzes the three architectures from public cases and lists what clinics should check when choosing.

**Key answer:** The key difference between counting systems is the record source: voice-based counting records what the physician says aloud, wearable sensing estimates insertion motions with machine learning, and per-needle identification gives every single needle its own identity. If records do not correspond to physical needles, verification has no cross-check — when the count falls short, you cannot tell a recording error from a real leftover. With per-needle identity, verification is a direct match between physical needles and records, and a shortfall can be pinned to a specific needle for further locating.

Source: https://www.program-world.com/en/insights/needle-counting-architectures/
Published: 2026-10-09
Tags: Smart acupuncture, Needle safety, Hospital cases, Selection guide

## How does Dalin Tzu Chi's voice-based counting system work?

The Department of Chinese Medicine at Dalin Tzu Chi Hospital and a team from National Chung Cheng University's computer science graduate institute jointly developed a smart acupuncture medical system, reported by Da Ai TV and Economic Daily News. The physician speaks the acupoints and needle counts while inserting; the system transcribes the speech in real time, renders the acupoints as graphics in the medical record and passes them back to the HIS. The team says this addresses unreliable physician counting, graphical acupoint records and record transfer. At removal, a staff member confirms on the system; the screen shows the remaining needle count, and zero means all needles are out.

The value of this system is that records are created in the moment: physicians no longer back-fill entries, which reduces omissions. As shown in the reports, the counting source is the physician's spoken dictation, accumulated after speech recognition. The correctness of acupoints and counts depends on the dictation and recognition quality, and records have no correspondence to the physical needles.

## Three counting architectures: where the record comes from

Public cases show three approaches. Dalin Tzu Chi records acupoints and counts from voice input. Taipei Veterans General Hospital's innovation project uses wrist-worn sensors to capture vibration and acceleration during insertion, with machine learning classifying valid insertion waveforms to count automatically — the project document notes the counting accuracy still requires further case collection for calibration. The third is per-needle physical identification: an identification element is integrated into the needle handle (patents held by our partner RFID Integrated Marketing), each needle gets its record at insertion, and the system matches physical needles against records at removal. This is the direction the Needle system takes; actual specifications are confirmed during site assessment.

The core difference is what a record corresponds to: dictation records what the physician said, sensor records are estimates from motion features, and per-needle identification corresponds to the physical needle itself. That determines the nature of verification — the first two verify internal consistency of records, the last matches physical needles against records.

| Architecture | Record source | Per-needle ID | Removal verification | Clues on a shortfall |
| --- | --- | --- | --- | --- |
| Voice counting (Dalin Tzu Chi) | Physician dictation + speech recognition | No | Staff confirms on system; screen shows remaining count | Trace dictation and recognition; no physical cross-check |
| Wearable sensing (Taipei VGHA) | Wrist sensors + machine-learning estimate | No | System counts automatically; accuracy requires further case collection | Re-trace sensed waveforms and classifications |
| Per-needle identification (commercial direction) | Identification element in needle handle | Yes | System matches physical needles against records and alerts | Pinpoints unverified needles; identification device assists locating |

## Why the record source drives verification and traceability

When records come from dictation or motion estimation, there is no cross-check between records and physical needles: if the count falls short after removal, you cannot distinguish an initial recording error (a skipped phrase, a mis-recognition) from a needle actually left on the patient. The system offers an audit trail of records, not physical locations. Under per-needle identification, each needle's identity is established at insertion; needles left unmatched are flagged at verification, and when something is missing an identification device can help search the treatment area (the Needle website's flow demo includes this segment; actual performance is subject to field assessment).

This is not to dismiss hospital-built systems — each solved its own pain point: voice recording reduces omissions and back-filling, wearable sensing tries to automate counting, and both are in-house projects with no sales information in the reports. For clinics and hospital TCM departments, the point is to know your own need: if the pain is digitizing records, voice recording already helps; if you need physical verification and leftover traceability, you need per-needle identification.

## What to check when evaluating

When evaluating a counting system, confirm four things in order. First, the record source — voice, sensing or physical identification, and the accuracy conditions each depends on. Second, the verification mode — manual confirmation on the system, or automatic matching with alerts. Third, exception handling — what clues the system gives when the count falls short, and whether it can assist locating. Fourth, integration and cost — HIS interfacing, needle specifications and deployment effort. Hospital cases serve as process references; this article is selection guidance, not medical advice, and actual deployment scope and cost are subject to site assessment.

## Where do voice counting and per-needle identification differ in record reliability?

Voice counting is only as correct as the dictation and speech recognition, and records have no cross-check against physical needles; per-needle identification gives each needle a standalone identity, so verification directly matches physical needles against records, and a shortfall can be pinned to a specific needle for locating.

## Can clinics adopt the hospitals' smart acupuncture systems directly?

The public cases are hospital-developed or academia-industry projects, and the reports carry no sales information; clinics can evaluate commercial solutions directly, comparing record source, verification mode and exception handling.

## What can a system do when the needle count falls short?

It depends on the architecture: record trails allow retracing dictation or recognition results; per-needle identification flags unmatched needles and assists locating with an identification device. System reminders assist care operations and do not replace the clinical judgment and verification duty of medical staff.

## Sources

- [Da Ai TV | report on the Dalin Tzu Chi and Chung Cheng University smart acupuncture system](https://www.daai.tv/news/taiwan/590623)
- [Economic Daily News | Dalin Tzu Chi Hospital smart acupuncture system](https://udn.com/news/story/7266/9184586)
- [Taipei VGHA smart cloud insertion/removal counting system (project document)](https://wd.vghtpe.gov.tw/cic/files/%E6%99%BA%E6%85%A7%E9%9B%B2%E7%AB%AF%E6%96%BD%E6%8B%94%E9%87%9D%E8%A8%88%E6%95%B8%E7%B3%BB%E7%B5%B1.pdf)
- [Taipei City Hospital acupuncture digitization report (Medicine journal study)](https://tw.news.yahoo.com/%E6%BC%8F%E6%8B%94%E9%81%9D-%E9%A2%A8%E9%9A%AA%E5%A4%A7%E9%99%8D%E5%85%AD%E6%88%90-%E5%8C%97%E5%B8%82%E8%81%AF%E9%86%AB%E9%87%9D%E7%81%B8%E5%B0%8E%E5%85%A5%E6%95%B8%E4%BD%8D%E7%A7%91%E6%8A%80%E6%88%90%E6%9E%9C%E7%99%BB%E5%9C%8B%E9%9A%9B%E6%9C%9F%E5%88%8A-051944214.html)