Typing is one of the quiet drains on a pathologist's day, and transcription queues can hold up reports that are otherwise ready. Voice dictation and AI scribes promise to close that gap. This article is for pathologists, pathology assistants, lab managers and IT teams considering voice tools at the grossing bench and at sign-out, with practical advice on setup, accuracy and governance.
From transcriptionists to speech recognition to AI scribes
Pathology has long relied on dictation. Traditionally, a pathologist or assistant recorded audio and a transcriptionist typed it, which added a delay and a second person to every report.
Speech recognition changed that by converting speech to text in real time, letting the speaker correct the text immediately. The newer generation, often called AI scribes, goes a step further: it combines speech recognition with a language model that can format dictation into report sections, apply templates and tidy phrasing. For the opportunities and risks of that language layer, see large language models and pathology reports.
Where voice dictation helps in the pathology lab
At the grossing bench
Grossing is a natural fit. Hands are gloved and busy, contamination must be avoided, and the description is produced while looking at the tissue. Voice lets the prosector describe dimensions, findings and block designations as they work, without touching a keyboard. Foot pedals or hands-free microphones avoid contact with shared equipment.
At sign-out
Pathologists can dictate microscopic descriptions, diagnoses and comments while still looking at the slide or the screen. Combined with templates, short voice commands can insert standard text, leaving the pathologist to dictate only what is specific to the case. See using report templates without losing clinical nuance for how to keep that balance.
For addenda and amendments
Quick dictation lowers the effort of adding an addendum when ancillary results arrive, which can encourage timely updates rather than batching them at the end of the day.
In multilingual settings
Many labs, including in Thailand, write reports in English while staff speak another language day to day. Check how the system copes with accents, mixed-language dictation and local names. A short trial with your own staff reading real, de-identified descriptions is far more informative than a vendor demonstration.
Accuracy: what to expect and how to improve it
Speech recognition has improved greatly, but pathology vocabulary is demanding: long Latin and Greek terms, eponyms, measurements, abbreviations and stain names. Errors that are minor in everyday speech can be serious in a report, such as confusing left and right, misheard numbers, or a missing "no".
Practical ways to improve accuracy:
- Use a medical or pathology vocabulary. General consumer speech recognition is not designed for clinical terms. Check whether the system can learn your lab's terminology and frequently used phrases.
- Invest in microphones. A good-quality, noise-cancelling microphone matters, especially in noisy grossing rooms with ventilation and equipment running.
- Train users and the system. Brief training on dictation habits, such as speaking punctuation and pausing between sections, pays off. Some systems improve with user-specific adaptation.
- Standardize phrasing. Consistent descriptions and templates make recognition more reliable and reports easier to read.
- Read before you sign. Every report must be proofread by its author. Pay particular attention to numbers, laterality, negations and units.
It also helps to agree on conventions for dictating numbers and units, for example always stating centimeters or millimeters explicitly and reading measurements in a fixed order. Small habits like these remove many avoidable corrections.
AI scribes: added value and added risk
An AI scribe can turn a stream of dictation into structured sections, apply formatting and suggest standard wording. That saves time, but it introduces a new risk: the system may rephrase what was said, merge or reorder statements, or drop a qualifier. In pathology, "consistent with" and "suspicious for" are not interchangeable.
Good practice includes keeping the original dictation available for comparison, highlighting text the system changed or added, and never allowing the scribe to introduce findings the pathologist did not dictate. As with any assistance, the pathologist reviews and signs the final report.
Privacy, security and governance
Voice recordings and transcripts contain patient information and must be protected like any other part of the record.
- Know where audio goes. Ask whether audio is processed on site or in the cloud, in which country, and whether it is stored or used to train models.
- Control access. Recordings and drafts should follow the same role-based access rules as reports.
- Secure the connection. Data in transit and at rest should be encrypted, and user accounts should be protected with strong authentication.
- Set retention rules. Decide how long audio is kept, and delete it when it is no longer needed.
- Keep an audit trail. Record who dictated, edited and signed each report.
Policies should also cover dictation in shared spaces, where other patients' details may be overheard or picked up by a nearby microphone.
Our lab manager's checklist for cloud security covers broader questions to ask any cloud provider.
Rolling out voice dictation
Start with a small group of willing users, ideally covering both grossing and sign-out. Measure what matters: time from grossing to completed description, time from slide review to final report, and the number of corrections found at proofreading. Gather feedback on microphones, commands and templates, then adjust before wider rollout.
Consider ergonomics too. Comfortable headsets, sensible microphone placement at the grossing station and a quiet place for sign-out dictation often matter more to adoption than feature lists.
Common mistakes include deploying without a pathology vocabulary, skipping microphone testing in the actual grossing room, and assuming accuracy is good enough to skip proofreading. Another is forgetting the people whose jobs change: transcription staff often know the lab's reports better than anyone and can become valuable editors, template owners or trainers.
Key takeaways
- Voice dictation removes typing and transcription delays at both the grossing bench and sign-out.
- Accuracy depends on vocabulary, microphones, user habits and standardized phrasing.
- AI scribes add formatting and structure but can alter meaning, so changes must be visible and reviewed.
- Treat audio and transcripts as patient data, with access control, encryption, retention rules and audit trails.
- Pilot with a small group and measure time saved and corrections needed.
PATHOLOGY+ includes voice dictation alongside report templates and AI-assisted drafting that the pathologist reviews and signs. Book a demo to hear it in action.