Gains:
- Ability to carry out end-to-end screening, diagnostic testing, device selection, fitting and follow-up stages on a single case with artificial intelligence support
- Ability to map which decision belongs to artificial intelligence and which belongs to the expert and verification points at each stage
- Ability to control workflow with an end-to-end quality and privacy checklist
In this final unit, we will combine everything you learned throughout the module into one patient journey. The aim is for you to see how to use artificial intelligence in the right role at every stage of the audiology workflow and to clearly answer the question "which decision belongs to artificial intelligence and which one belongs to the expert" at each stage. The main principle is once again: AI is an assistant, pre-screener and draft generator through and through; Diagnosis, device prescription, fitting decision, clinical interpretation and all safety-critical decisions belong to the specialist and, when necessary, the physician after verification.
Our Case (anonymous)
A 62-year-old male patient applies with the complaint that "for the last year, I have been having difficulty understanding conversations, especially in crowded environments, and I turn up the volume of the television too much." He worked in a noisy production environment for many years before retirement. We will follow this patient from screening/pre-assessment to follow-up. (All data is kept in the enterprise system; only the anonymous summary goes to the AI.)
Step by step workflow
1. Anamnesis and preliminary evaluation
The audiologist asks the artificial intelligence for a list of history questions appropriate to this profile (noise exposure, medication, family history, tinnitus, balance). AI role: list of reminder questions. Expert judgment: which questions to apply to this patient and the interpretation of the answers.
2. Audiogram and immittance
Tests are performed: high-frequency predominant, symmetrical, moderate sensorineural loss on the right and left; Type A tympanogram in both ears. The audiologist gives the thresholds to the artificial intelligence in an anonymous and structured form and receives a type-degree outline. AI role: pre-interpretation draft. Expert judgment: verification by calibrated measurement, checking the need for masking, final interpretation.
Hint: Since both ears are symmetrical in this patient, the masking dilemma is limited; but it should be a habit to do your own reading and verify rather than blindly trusting the AI draft.
3. Tinnitus/balance red flag screening
The patient describes mild, bilateral, persistent tinnitus; no sudden hearing loss, unilateralism, pulsatile features, or dizziness. The audiologist runs the red flag filter; There is no emergency sign. AI role: sorting red flag questions. Expert judgment: urgency assessment and management.
4. Device selection
The patient is active, uses the phone a lot, has good manual dexterity, medium budget. The audiologist orders a brand-free comparison chart with an anonymous profile. AI role: type/feature comparison and realistic expectation presentation. Expert judgment: final device prescription.
5.Fitting and REM
The device is programmed with target NAL-NL2; After the first adjustment, the patient says, "My own voice is a bit strange." The audiologist translates the complaint to artificial intelligence and gets possible directions, then measures REM and actual output against the target and corrects the setting. AI role: translating the complaint into setting language. Expert judgment: REM measurement and final setting.
6. Patient education
The audiologist has the patient receive a personalized, plain-language "first two weeks getting used to it" brochure from the artificial intelligence; Corrects exaggerated promise sentences. AI role: draft material. Expert judgment: accuracy and confirmation.
7. Report
The findings are compiled into a manuscript with the prohibition of fabrication rule; The audiologist compares each line to the source data and signs it. AI role: draft report. Expert judgment: verification and signature.
8. Follow-up (teleaudiology)
After one month, anonymous usage data from the device app (~6 hours per day, mostly noisy environment) is summarized by the AI; The patient says, "I'm still having a hard time at the restaurant." The audiologist does not implement the recommendation directly, but discusses the setting with the patient and approves it. AI role: trend summary. Expert judgment: setting confirmation.
decision map
Stage
Role of AI
Expert decision
verification point
anamnesis
Question list
Comment
clinical strainer
audiogram
Pre-commentary draft
Type/degree
Calibrated measurement
red flag
Question sorting
urgency
Clinical evaluation
Device selection
Comparison
prescription
Clinic + patient need
Fitting
Complaint translation
Last setting
REM
Education
Material draft
Approval
Readability + accuracy
Report
draft
signature
Source comparison
tracking
Trend summary
Setting confirmation
Patient + measurement if necessary
Weak prompt / Strong prompt
Weak prompt:
Manage the entire process of this patient: diagnose, select device, make setting, write report, finish.
It delegates all end-to-end clinical decisions to artificial intelligence; it completely bypasses validation, measurement, and expert validation.
Powerful prompt:
Your role: end-to-end process assistant. Do not make a diagnosis/prescription/fitting DECISION at any stage; For each step, (1) produce a sketch that you can help with, (2) indicate your decision and verification point. Case (anonymous): 62 years old, noise history, difficulty understanding in crowds. Task: anamnesis, pre-interpretation, red flag, device comparison, fitting troubleshooting, training, report, follow-up. List the assistant printout and verification step for each of the follow-up steps separately.
The strong will separates each stage, leaves the decision and verification to the expert, keeping the AI as a mere assistant.
Three mini cases (same patient, three critical moments)
Case moment 1 — The value of verification. In the audiogram outline, the AI classifies the loss as "mild"; When the expert recalculates the average, he sees that it is “medium” — the AI misweighted a few frequencies. This 10 dB difference affects the device technology level decision. Without verification, the wrong class would be reported.
Incident moment 2 — Hallucination. In the report draft, the AI adds “speech discrimination score 92%”; However, this test has not been performed. The expert catches this in the source comparison and writes "[no speaking test]". A single line captured before signature maintains the validity of the report.
Incident moment 3 — Privacy. In the follow-up phase, an assistant is about to enter the patient's name and device serial number for the AI summary; the rule is remembered and the data is anonymized as "62 years old, RIC device, ~6 hours per day". The clinical decision does not change, the patient is protected.
Copiable prompt templates
END-TO-END ROLE MAP TEMPLATEList the role of artificial intelligence and the expert's decision and verification point in a table at each stage (anamnesis, pre-interpretation, red flag, device, fitting, training, report, follow-up) for the following case. Don't decide, just map.Case (anonymous): [case]
PHASE TRANSITION CONTROL TEMPLATEI am completing the following phase: [phase]. Give a checklist of points that need to be verified (measurement, expert approval, source comparison) before moving on to the next stage.
QUALITY + PRIVACY AUDIT TEMPLATEAudit this end-to-end workflow under two headings: (1) whether there is verification and expert approval of each clinical decision, (2) whether identity data is entered at any stage. Mark the missing ones. Stream summary: [summary]
CASE SUMMARY ANONYMIZATION TEMPLATEFrom the case summary below, remove any elements that might identify the person and transform them into a clinically adequate, anonymous educational/presentation summary. Summary: [case]
Common mistakes
- Skip the stage. Passing the verification point and moving to the next stage early.
- Relying on a single output. Making the draft of artificial intelligence at one stage the basis for subsequent decisions without verification.
- Forgetting the decision map. Not to re-question at every stage which decision belongs to the expert.
- Relaxing privacy between stages. Neglecting anonymization in tracking or reporting.
- It means "manage from end to end". Writing a single command that hands over the entire process to the AI.
In summary
Audiology workflow; It is a chain that extends from anamnesis to follow-up, where artificial intelligence can help at every stage, but the decision belongs to the expert. Artificial intelligence saves time at every stage by drafting, reminding and pre-screening; However, diagnosis, prescription, fitting and clinical interpretation are left to the specialist and, when necessary, the physician after verification. Each stage has a verification point, a quality and privacy audit of the entire process. Unverified output is as invalid as an unsigned report.
Application task
Construct an anonymous case from your own practice and chart the role of the AI, the expert's decision, and the point of verification for each of the eight stages with an "end-to-end role map" template. Then audit your flow with the “quality + privacy audit” template: is verification missing somewhere, is identity data leaked somewhere? Write down in one sentence how you would close each gap you found.
checklist
- [ ] I separated the role of artificial intelligence and the decision of the expert at each stage.
- [ ] I defined a verification point for each stage.
- [ ] I verified critical outputs such as audiogram/report with the source/measurement.
- [ ] I ran the red flag scan before general operations.
- [ ] I kept patient data anonymous at all stages.
- [ ] I audited the end-to-end stream for quality and privacy.
Module Exam
1. An audiologist transfers the audiogram interpretation produced by artificial intelligence to the patient's report without checking it. What is the fundamental mistake in this approach?
- A) Artificial intelligence output cannot be used without verification; Draft comments must be expertly reviewed and approved ✔
- B) Artificial intelligence always gives worse audiogram interpretation than it is
- C) Only numerical threshold values should be reported instead of comments
- D) The report should have been given to the patient on paper instead of e-mail.
Explanation: While artificial intelligence can produce a correct interpretation, it can also incorrectly assume the type, degree of hearing loss or the need for masking and produce an incorrect interpretation with the same confidence. Audiology is a safety-critical field; Each output must be calibrated and expertly verified, and diagnosis and final approval must be human.
2. The audiogram shows a significant gap between air conduction and bone conduction thresholds (air-bone gap). What does this finding typically suggest?
- A) There may be a conductive component (originating from the outer/middle ear).
- B) There is definitely permanent neural hearing loss
- C) It may be a conduction type component; However, clinical evaluation clarifies the type ✔
- D) It proves that the device's calibration is broken
Explanation: Worse airway thresholds while the bone conduction (inner ear/cochlea function) is close to normal suggests that there is a problem in the conduction of sound from the outer/middle ear, that is, there is a conductive component. AI may indicate this in the sketch, but the type distinction and its reason become clear through clinical evaluation.
3. In newborn hearing screening, the baby receives a 'pass' result from the OAE test. The family says, "This means my child will not have any hearing problems." What is the best approach?
- A) Yes, if the screening is passed, it is impossible to have hearing problems in the future
- B) Scan shows current status; Follow-up and symptom monitoring are important for losses that may develop later ✔
- C) The test may be incorrect, the device should be inserted immediately
- D) The scan result is unreliable and should not be taken into account
Explanation: Screening test is not a diagnosis; A 'pass' result indicates that there is no significant problem with the current scan, and does not guarantee lifetime hearing. Some hearing losses may develop later or may be of a type that OAE cannot detect (e.g. auditory neuropathy). This limit should be explained to the family in plain language.
4. In tympanometry, a smooth curve (Type B) and normal ear canal volume are obtained. Artificial intelligence interprets this. What should be the expert's primary assessment?
- A) It may suggest a middle ear problem (e.g. fluid); combined with clinical examination and ENT if necessary ✔
- B) There is definitely a hole in the eardrum
- C) The result is normal, no action is required
- D) Hearing aid should be prescribed directly to the patient
Description: A flat (Type B) tympanogram with normal canal volume is suggestive of middle ear fluid or decreased middle ear motility and usually requires ENT evaluation. The AI blueprint is a start; Tympanogram alone does not make a diagnosis; it is combined with clinical examination and history.
5. An expert who consults artificial intelligence on hearing aid selection notices that the output consistently highlights a particular brand. What is the best attitude in this situation?
- A) That brand is definitely the best because it is recommended by artificial intelligence
- B) The most expensive device is always the best choice
- C) Device selection should be left entirely to artificial intelligence
- D) Brand orientation may be a data bias; The choice should be based on objective patient needs ✔
Explanation: AI output may reflect biases in the training data, and trademark guidance may be based on data bias rather than the patient's best interest. Device selection should be made according to objective criteria such as hearing loss profile, lifestyle, dexterity and budget; Patient need, not the brand, should be the determining factor.
6. During hearing aid programming (fitting), artificial intelligence recommends certain gain values. What is the most reliable way to validate this recommendation in the clinic?
- A) Ask the patient 'Is he okay?' It is enough to ask
- B) Trust and save the recommendation of artificial intelligence
- C) Verify whether the prescription target has been achieved with real ear measurement (REM) ✔
- D) Preserving the factory settings in the box of the device
Description: Real-ear measurement (REM, real-ear measurement) measures the actual output produced by the device in the patient's own ear canal and indicates whether the prescription target (e.g. NAL-NL2) has been achieved. The default values of the AI or software are not a substitute for real ear acoustics; verification is done with REM.
7. A patient describes new-onset, rapidly increasing hearing loss and unilateral tinnitus in one ear. AI suggests general 'tinnitus management' material. What should be the expert's priority?
- A) Give general tinnitus material and send the patient home
- B) Trying to sell hearing aids directly
- C) Recognizing that this may be a red flag that requires urgent referral and directing the patient to the physician without delay ✔
- D) Suggesting to wait a few months and let it go away on its own
Description: Sudden/unilateral hearing loss and unilateral tinnitus are red flags that require immediate medical evaluation (e.g., sudden sensorineural hearing loss is a condition with an early treatment window). The general context of AI should not overshadow this urgency; The patient should be referred to a physician without delay.
8. An audiologist uses artificial intelligence to create a plain-language educational brochure for the patient. In the material, artificial intelligence states, "This device completely returns your hearing to normal." What is the correct fix?
- A) The sentence is correct, it should be left as is.
- B) The sentence creates exaggerated and unrealistic expectations; should be translated into realistic 'helps heal' language ✔
- C) Even stronger promises should be added so that the patient is convinced
- D) The brochure should be written entirely in technical jargon.
Explanation: Hearing aids do not 'fully normalize' hearing; Helps significantly improve hearing and communication. Exaggerated promises create unrealistic expectations in the patient and are unethical. The material should be rewritten in realistic and honest language.
9. When drafting an audiological report, artificial intelligence fabricates an 'acoustic reflex result' that the expert did not give. What is this situation called and how can it be prevented?
- A) Hallucination; Prevented by 'write only from the data I provide, mark [no data] if missing' instruction and source check before signing ✔
- B) Calibration error; Prevented by restarting the device
- C) Normal behavior; The report can be taken verbatim
- D) Network problem; Prevented by correcting the internet connection
Explanation: When artificial intelligence produces a finding that is not given, it is called a hallucination. His precaution is to explicitly request that the model only write from the actual measurement data provided, mark missing data as '[no data]', and not make predictions/fitting. Each report should be compared to the source data before signature.
10. An audiologist pastes the patient's name, ID number, and full audiogram into a public cloud AI tool and requests comments. What is the main problem with this behavior?
- A) No problem, artificial intelligence keeps the data private
- B) He should have sent the entire file, not just the audiogram
- C) The only problem is that the comment comes slowly
- D) Identity data has been sent to the public tool; This is a violation of KVKK, data should be anonymized and safe vehicles should be used ✔
Explanation: Sending data that directly identifies the person (name, TR ID number) to a public tool is a serious violation in terms of KVKK and patient privacy. The correct approach is to anonymize data (remove identifying information), use corporate/secure tools, and share only necessary clinical data.
11. In teleaudiology (remote monitoring), artificial intelligence produces adjustment recommendations based on usage data from the patient's hearing aid application. What should be done before implementing this recommendation?
- A) The recommendation should be applied to the device immediately and automatically
- B) The recommendation should be confirmed by the specialist's clinical evaluation and approved together with the patient complaint and profile ✔
- C) Usage data is unreliable and should be completely ignored
- D) The patient himself needs to change the setting with artificial intelligence
Explanation: Remote usage data is a valuable clue, but the setting decision requires clinical judgment from the expert. The recommendation should be confirmed by the patient's actual complaint, hearing profile and, if necessary, new measurements; The device setting should not be changed permanently without expert approval.
12. Which of the following is the most accurate general framework for the use of artificial intelligence in audiology?
- A) Artificial intelligence can make all decisions including diagnosis and prescription, experts are unnecessary
- B) Artificial intelligence is of no use in audiology and should not be used
- C) If the artificial intelligence output is correct, even patient consent is unnecessary
- D) Artificial intelligence is an assistant and draft generator; Diagnosis, prescription and fitting decisions belong to the specialist after verification ✔
Description: Artificial intelligence saves time in audiology as an assistant, pre-screener and sketch generator; However, diagnosis, device prescription, fitting decision and clinical interpretation belong to the competent specialist. Outputs should be verified by calibrated measurement and clinical evaluation, and human approval should be essential in safety-critical decisions.
13. When is masking required in pure tone audiometry and what is the role of artificial intelligence here?
- A) Masking is never required; artificial intelligence determines this
- B) Masking is required only in children
- C) Masking is a process performed entirely automatically by artificial intelligence.
- D) It is necessary if the difference between the ears is large enough to cause crossed hearing; AI may be a reminder, but it's up to the expert to decide ✔
Explanation: Masking is required when the difference between the thresholds of the two ears is large enough to cause sound to pass across the skull to the untested ear (cross-hearing); otherwise the threshold may belong to the wrong ear. The AI may be a control assistant that reminds you of the need to mask, but the decision and implementation of masking is based on the expert's calibration test.
14. In workplace noise scanning, artificial intelligence quickly classifies workers' results and flags those that need follow-up. What is the correct limit of this usage?
- A) Artificial intelligence helps with follow-up prioritization; However, screening is not diagnosis and diagnostic guidance and responsibilities belong to the specialist ✔
- B) Artificial intelligence can directly diagnose workers and sign reports
- C) If the scan results are confirmed by artificial intelligence, no verification is required
- D) Privacy rules have no importance in workplace screening
Description: AI is a useful pre-screener that prioritizes those that require follow-up in screening lists; But screening is not diagnostic and cannot completely prevent false negatives (missed cases). Directing workers with hearing loss for diagnostic evaluation, notification and protection obligations are the responsibility of experts and occupational health processes.