Reporting and Dictation

Abstract

Since the discovery of the X-ray in 1896 radiology reporting has changed very little until the past decade. Very early reports were typed or handwritten by the radiologist. Later, reports were dictated into a recording device and later transcribed to text. Most reports were in prose format. Fewer were in a list or itemized configuration. There has been recent renewed interest in other reporting techniques. Speech recognition has been available for a number of years and has become more prevalent, albeit controversial, in the past decade. Although structured reporting has been available for a number of years, it is not widespread in radiology practice.

Self-Assessment Questions

• 1. All of the following pertain to the use of macros and templates in speech recognition except

  • a. they are essential for efficient use

  • b. users should be encouraged to create their own macros

  • c. should be kept to few in number to avoid confusion

  • d. are most effective for users with poor recognition

• 2. When troubleshooting a radiologist complaining of poor accuracy in speech recognition you should do everything except

  • a. check the microphone connection with the soundcard

  • b. encourage the radiologist to speak more slowly

  • c. check soundcard quality using speech engine software

  • d. encourage the radiologist to use a headset microphone

• 3. A point-and-click structured report with tracked data elements

  • a. is used predominantly in general radiology

  • b. can improve radiologist efficiency and decrease distraction during image interpretation

  • c. is ideal for data mining

  • d. is better accepted by radiologists than is speech recognition

• 4. All of the following are true concerning an itemized report except

  • a. it is preferred by clinicians over a conventional report

  • b. it saves time both in report creation and report interpretation

  • c. it is not common in radiology reporting

  • d. it is a form of structured reporting

• 5. Concerning interoperability of speech recognition and structured reporting with other software such as PACS and RIS

  • a. it is not necessary in smaller hospitals and imaging centers

  • b. it improves mainly clerical but not clinical workflow

  • c. integration is now standardized across all PACS and reporting vendors

  • d. it may require an upgrade to workstation hardware

• 6. RadLex

  • a. is a lexicon created by a single individual

  • b. content was not dependent on existing lexicons

  • c. decreases ambiguity in reports

  • d. is simple and easy to manage

Definition 10.1: Speech Recognition

A software application that automatically converts spoken words into text. SR can replace a medical transcriptionist.

Definition 10.2: Turnaround Time

The time interval between the completion of a study and the study’s signed and finalized report becoming available in the medical record.

Key Concept 10.3: Speech Recognition Engine

SR software is conceptually divided into an “engine” that performs the actual conversion from speech to text, and an “interface” that provides workflow and interacts with other software applications.

Key Concept 10.4: Implementing SR

User buy-in and a project champion (preferably a radiology department leader) should be secured before any decision to implement speech recognition.

Key Concept 10.5: Addenda

Once a radiology report has been finalized by the radiologist, it cannot be modified because it is a permanent part of the medical record. Changes and clarifications must be added to the end of the report without changing the original report. SR software must support this workflow.

Step-by-Step 10.6: Vendor Evaluation

• 1. Assess speech engine accuracy, keeping in mind the limitations of short-term evaluation.

• 2. Observe multiple levels of users (residents, staff, experienced as well as struggling users) at a site visit.

• 3. Evaluate level of integration with PACS and RIS specifically as it pertains to your own software.

• 4. Assess radiologist workflow specifically pertaining to navigation of text while eyes are on PACS images.

• 5. Ensure both initial and ongoing application support. Will you or the vendor be training new users and supporting current ones?

• 6. Evaluate upgrade path and roadmap of future enhancements.

• 7. Observe how dual reads, resident workflow, and addenda are handled.

Checklist 10.7: Training Issues to Consider

• Expect a longer learning curve for part-timers and those radiologists moving from site to site with different reporting systems.

• If you plan to employ locum tenens you must accommodate this workflow.

• The software can accommodate many, but not all, heavy accents. If you have nonnative English speakers, plan for transcriptionist correction or alternative dictation.

Definition 10.8: Self-Edit Mode

Users dictate, edit, and sign reports without the aid of backend human transcription. Reports can be completed one at a time or batched.

Definition 10.9: Transcriptionist Mode

Users dictate in a conventional manner, and SR is applied. The text and wav file are then sent to a transcriptionist where corrections are made. A corrected draft report is returned to radiologist for final approval.

Key Concept 10.10: Batch Mode

A radiologist may sign each report as it is dictated, or may dictate several reports and then sign them all in a batch.

Our Experience 10.11: PEARLS from an SR Administrator

• Try to be involved in speech recognition as soon as you are assigned. It is not too early to sit in on vendor evaluation discussions.

• Shadow the vendor applications’ specialists and learn as much as you can about the software and its use during installation and training.

• Attend user-group meetings and make contact with other administrators.

• Be as visible as possible at all times within your department. If possible, make regular “rounds” and actively ask users if there are problems or issues that need attention.

• Try to sit as much as you can with the best radiologist users to learn tips on use and efficiency.

• Learn all you can about macros and templates. They are the key to user efficiency.

Source: Gretchen Oman, RT, Imaging Informatics IT specialist in charge of speech recognition

Definition 10.12: Macros and Templates

Predefined reports, words, phrases, or other data that are used to shorten reporting time. These can consist of an entire report and often are created with blanks that can be filled in with other data such as numerical values. Some distinguish between a macro, a complete stored report versus a template, a report created with blank fields for the easy insertion of other data elements, numerical, or otherwise.

Our Experience 10.13: User Acceptance

In a small community hospital of 100 beds performing 100,000 exams per year, the full staff of four radiologists agreed to implement speech recognition. Within one week, all radiologists were using speech recognition and no reports were sent to transcriptionists. Within four weeks, workflow and productivity were subjectively back to baseline levels.

In a larger more diverse department, the decision was made to implement speech recognition. Radiologists were encouraged but not required to use speech recognition. After six months, approximately 50% of radiologists had adapted speech recognition voluntarily. The remainder continued to use conventional dictation.

Key Lessons Learned

• Get as much initial buy-in as possible.

• Consider mandating use of speech recognition.

• Allow back-end transcription but incentivize use of self editing.

• Every term paper has a due date. Set an end point for removing manual dictation.

• In a large and diverse department, consider implementing in stages.

Key Concept 10.14: Microphone Position

Faulty microphone position is one of the most common causes of poor accuracy, even among experienced users. This variable is removed nearly completely if a headset microphone is used.

Step-by-Step 10.15: Troubleshooting Poor Accuracy

• 1. The first task when called to a problem workstation is to check that the microphone is connected properly.

• 2. Next, test microphone and soundcard quality using the speech engine software. If this is unsatisfactory, change the microphone. If this does not work, the soundcard may be faulty.

• 3. Make sure that extraneous sounds are not interfering with accuracy. The worst of these are sudden loud noises such as doors closing or an overhead intercom.

• 4. Carpeting, acoustic ceiling tile, and sometimes acoustic wall panels are helpful. Consider the use of a white noise generator for problem areas.

Key Concept 10.16:Language Models

As it is used, the SR software learns the speech pattern, vocabulary, and accent of each individual user. This is called a personal language model. A radiologist who tries to dictate under someone else’s login will be unpleasantly surprised by the result! The language model can sometimes be modified manually using a vocabulary editor.

Hypothetical Scenario 10.17

A user tries to make a correction and insert the word “and” for the mistaken text “an.” These are near-homophones and the speech engine will likely not sense the slight difference in pronunciation. It is better to correct the phrase around the offending word to give more data and context. This will improve accuracy in dictation and especially in correction.

Definition 10.18: Structured Reporting

There is some confusion as to the exact definition of structured reporting. According to Dr. Curt Langlotz, it can be considered in a three-pronged approach with three different features.

• Feature 1. A report that is consistent throughout with headings such as HISTORY, FINDING, and IMPRESSION.

• Feature 2. A report that lists organ systems in an organized itemized fashion with prose or shorter descriptive terms following each heading. This type of report is sometimes called an itemized report.

• Feature 3. A report that uses a standard lexicon with all findings codified within the database for billing, indexed search, and data mining capability.

Key Concepts 10.19: Macros and Templates

• Macros and templates are a key feature of both speech recognition and structured reporting. Their use will determine in large part whether a user is efficient and accepting of the product.

• For speech recognition, the worse the recognition accuracy, the more important macros and templates become.

• A systematic naming system across all modalities will be helpful to the user as more macros are added. Many users ultimately have over 100 macros.

• Macros and templates should be created to be easily modified using only voice commands (for speech recognition). This will allow the user to remain focused on the images while creating a report.

• With newer systems, the appropriate macro can be automatically selected and presented to the user by using data from RIS and modality, such as type of study and patient demographics.

 Note that the brackets define fields that can be easily identified and modified.

Each organ system or related concept is a separate sentence. These sentences can be easily selected and modified using voice commands in either speech recognition or structured reporting. This keeps the eyes of the user on the images.

Any frequently used modification such as a one sentence description of fatty liver can be made into its own macro and inserted within this template.

Anything that a user is reporting more than once or twice a day should be made into a macro. This can be a full report or a part of a report.

Hypothetical Scenario 10.21: Off-the-Shelf Software

Dude, I can save some serious coin by purchasing an off-the-shelf speech engine and using this for dictation. What is wrong with this strategy?

For a single radiologist without a RIS or PACS this is a marginally acceptable solution to improve turnaround time and save money on transcription. The speech engine could be trained for radiology report structure and a medical vocabulary could be purchased or added manually. Newer speech engine software will even allow multiple users and has increasingly sophisticated macro capability. What you are paying the reporting vendor for is mainly interoperability and workflow enhancements.

Definition 10.22: Lexicon

A related set or collection of terms, labels, or entities.

Definition 10.23: Ontology

A domain that formally defines a set of classes of terms (“entities”), attributes of those terms (“slots”), and relationships of the terms.

Pearls 10.24

• Remember to evaluate workflow functionality and not just report creation for both speech recognition and structured reporting.

• Take steps to maximize user accuracy in speech recognition including microphone position and proper dictation techniques.

• Plan on at least temporary productivity decreases when implementing speech recognition or structured reporting.

• Consider structured reporting for limited use until newer software is more mature.

• Encourage the use of macros and templates in both speech recognition and structured reporting.

• Carefully plan integration and interface strategies specific to your PACS and RIS.

• Become familiar with the RadLex project as a facilitator of structured reporting.