From manual to artificial intelligence fitting: Two cochlear implant case studies

Artificial Intelligence in Otology and Neurotology

Clinical applications of artificial intelligence (AI) have grown exponentially with increasing computational power and Big Data. Data rich fields such as Otology and Neurotology are still in the infancy of harnessing the power of AI but are increasingly involved in training and developing ways to incorporate AI into patient care. Current studies involving AI are focused on accessible datasets; health care wearables, tabular data from electronic medical records, electrophysiologic measurements, imaging, and "omics" provide huge amounts of data to utilize. Health care wearables, such as hearing aids and cochlear implants, are a ripe environment for AI implementation.

Preliminary Evidence of the Use of Generative AI in Health Care Clinical Services: Systematic Narrative Review

BACKGROUND

Generative artificial intelligence tools and applications (GenAI) are being increasingly used in health care. Physicians, specialists, and other providers have started primarily using GenAI as an aid or tool to gather knowledge, provide information, train, or generate suggestive dialogue between physicians and patients or between physicians and patients' families or friends. However, unless the use of GenAI is oriented to be helpful in clinical service encounters that can improve the accuracy of diagnosis, treatment, and patient outcomes, the expected potential will not be achieved. As adoption continues, it is essential to validate the effectiveness of the infusion of GenAI as an intelligent technology in service encounters to understand the gap in actual clinical service use of GenAI.

OBJECTIVE

This study synthesizes preliminary evidence on how GenAI assists, guides, and automates clinical service rendering and encounters in health care The review scope was limited to articles published in peer-reviewed medical journals.

METHODS

We screened and selected 0.38% (161/42,459) of articles published between January 1, 2020, and May 31, 2023, identified from PubMed. We followed the protocols outlined in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines to select highly relevant studies with at least 1 element on clinical use, evaluation, and validation to provide evidence of GenAI use in clinical services. The articles were classified based on their relevance to clinical service functions or activities using the descriptive and analytical information presented in the articles.

RESULTS

Of 161 articles, 141 (87.6%) reported using GenAI to assist services through knowledge access, collation, and filtering. GenAI was used for disease detection (19/161, 11.8%), diagnosis (14/161, 8.7%), and screening processes (12/161, 7.5%) in the areas of radiology (17/161, 10.6%), cardiology (12/161, 7.5%), gastrointestinal medicine (4/161, 2.5%), and diabetes (6/161, 3.7%). The literature synthesis in this study suggests that GenAI is mainly used for diagnostic processes, improvement of diagnosis accuracy, and screening and diagnostic purposes using knowledge access. Although this solves the problem of knowledge access and may improve diagnostic accuracy, it is oriented toward higher value creation in health care.

CONCLUSIONS

GenAI informs rather than assisting or automating clinical service functions in health care. There is potential in clinical service, but it has yet to be actualized for GenAI. More clinical service-level evidence that GenAI is used to streamline some functions or provides more automated help than only information retrieval is needed. To transform health care as purported, more studies related to GenAI applications must automate and guide human-performed services and keep up with the optimism that forward-thinking health care organizations will take advantage of GenAI.

Prospective Comparison Between Manual and Computer-Assisted (FOX) Cochlear Implant Fitting in Newly Implanted Patients

OBJECTIVE

A prospective, longitudinal, randomized controlled trial with an original crossover design for 1 year was conducted to compare manual fitting to artificial intelligence-based fitting in newly implanted patients.

DESIGN

Twenty-four patients who received their first cochlear implant (CI) were randomly assigned to the manual or Fitting to Outcome eXpert (FOX) arm; they followed the corresponding fitting procedures for 1 year. After 1 year, each patient was switched to another arm. The number of fittings, auditory outcomes (pure-tone thresholds, loudness scaling curves, spectral discrimination scores, bisyllabic word recognition in quiet and noise, and speech tracking), fitting session questionnaire, and CI parameters (T level, C level, Threshold Sound Pressure Level (T-SPL), Comfortable Sound Pressure Level (C-SPL), and loudness growth value) were compared between the two groups. Differences between the two groups were analyzed using the Mann-Whitney test, and Holm corrections were applied for multiple statistical tests. At the end of the crossover session, patients were offered the choice to continue with their old or new map.

RESULTS

As early as 3 mo postactivation, the FOX group showed less variability and significantly better speech intelligibility in quiet conditions at 40 and 55 dB SPL and noise ( p < 0.05) with median phoneme scores of 50%, 70%, and 50% at 55, 70, and 85 dB SPL compared with 45%, 50%, and 40%, respectively. This group showed better results at 12 mo postactivation ( p < 0.05). In the manual group, 100% of the patients decided to keep the new FOX map, and 82% performed better with the FOX map. In the FOX group, 63% of the patients decided to keep the manual map, although the measurable outcome had not improved. In this group, participants reported to prefer the manual map because it felt more comfortable, even if the FOX map gave better measured outcome.

CONCLUSION

Although the study size remains relatively small, the AI-FOX approach was equivalent to or even outperformed the manual approach in hearing performance, comfort, and resources. Furthermore, FOX is a tool capable of continuous improvement by comparing its predictions with observed results and is continuously learning from clinicians' practice, which is why this technology promises major advances in the future.

Dichotic spectral integration range for consonant recognition in listeners with normal hearing

Dichotic spectral integration range, or DSIR, was measured for consonant recognition with normal-hearing listeners. DSIR is defined as a frequency range needed from 0 to 8,000 Hz band in one ear for consonant recognition when low-frequency information of the same consonant was presented to the opposite ear. DSIR was measured under the three signal processing conditions: (1) unprocessed, (2) target: intensified target spectro-temporal regions by 6 dB responsible for consonant recognition, and (3) target minus conflicting: intensified target regions minus spectro-temporal regions that increase confusion. Each consonant was low-pass filtered with a cutoff frequency of 250, 500, 750, and 1,000 Hz, and then was presented in the left ear or low-frequency (LF) ear. To create dichotic listening, the same consonant was simultaneously presented to the right ear or high-frequency (HF) ear. This was high-pass filtered with an initial cutoff frequency of 7,000 Hz, which was adjusted using an adaptive procedure to find the maximum high-pass cutoff for 99.99% correct consonant recognition. Mean DSIRs spanned from 3,198-8,000 Hz to 4,668-8,000 Hz (i.e., mid-to-high frequencies were unnecessary), depending on low-frequency information in the LF ear. DSIRs narrowed (i.e., required less frequency information) with increasing low-frequency information in the LF ear. However, the mean DSIRs were not significantly affected by the signal processing except at the low-pass cutoff frequency of 250 Hz. The individual consonant analyses revealed that /ta/, /da/, /sa/, and /za/ required the smallest DSIR, while /ka/, /ga/, /fa/, and /va/ required the largest DSIRs. DSIRs also narrowed with increasing low-frequency information for the two signal processing conditions except for 250 vs. 1,000 Hz under the target-conflicting condition. The results suggest that consonant recognition is possible with large amounts of spectral information missing if complementary spectral information is integrated across ears. DSIR is consonant-specific and relatively consistent, regardless of signal processing. The results will help determine the minimum spectral range needed in one ear for consonant recognition if limited low spectral information is available in the opposite ear.

Recent advancements in bioelectronic devices to interface with the peripheral vestibular system

Balance disorders affect approximately 30% of the population throughout their lives and result in debilitating symptoms, such as spontaneous vertigo, nystagmus, and oscillopsia. The main cause of balance disorders is peripheral vestibular dysfunction, which may occur as a result of hair cell loss, neural dysfunction, or mechanical (and morphological) abnormality. The most common cause of vestibular dysfunction is arguably vestibular hair cell damage, which can result from an array of factors, such as ototoxicity, trauma, genetics, and ageing. One promising therapy is the vestibular prosthesis, which leverages the success of the cochlear implant, and endeavours to electrically integrate the primary vestibular afferents with the vestibular scene. Other translational approaches of interest include stem cell regeneration and gene therapies, which aim to restore or modify inner ear receptor function. However, both of these techniques are in their infancy and are currently undergoing further characterization and development in the laboratory, using animal models. Another promising translational avenue to treating vestibular hair cell dysfunction is the potential development of artificial biocompatible hair cell sensors, aiming to replicate functional hair cells and generate synthetic 'receptor potentials' for sensory coding of vestibular stimuli to the brain. Recently, artificial hair cell sensors have demonstrated significant promise, with improvements in their output, such as sensitivity and frequency selectivity. This article reviews the history and current state of bioelectronic devices to interface with the labyrinth, spanning the vestibular implant and artificial hair cell sensors.

Effect of the Target and Conflicting Frequency and Time Ranges on Consonant Enhancement in Normal-Hearing Listeners

In this paper, the effects of intensifying useful frequency and time regions (target frequency and time ranges) and the removal of detrimental frequency and time regions (conflicting frequency and time ranges) for consonant enhancement were determined. Thirteen normal-hearing (NH) listeners participated in two experiments. In the first experiment, the target and conflicting frequency and time ranges for each consonant were identified under a quiet, dichotic listening condition by analyzing consonant confusion matrices. The target frequency range was defined as the frequency range that provided the highest performance and was decreased 40% from the peak performance from both high-pass filtering (HPF) and low-pass filtering (LPF) schemes. The conflicting frequency range was defined as the frequency range that yielded the peak errors of the most confused consonants and was 20% less than the peak error from both filtering schemes. The target time range was defined as a consonant segment that provided the highest performance and was decreased 40% from that peak performance when the duration of the consonant was systematically truncated from the onset. The conflicting time ranges were defined on the coincided target time range because, if they temporarily coincide, the conflicting frequency ranges would be the most detrimental factor affecting the target frequency ranges. In the second experiment, consonant recognition was binaurally measured in noise under three signal processing conditions: unprocessed, intensified target ranges by a 6-dB gain (target), and combined intensified target and removed conflicting ranges (target-conflicting). The results showed that consonant recognition improved significantly with the target condition but greatly deteriorated with a target-conflicting condition. The target condition helped transmit voicing and manner cues while the target-conflicting condition limited the transmission of these cues. Confusion analyses showed that the effect of the signal processing on consonant improvement was consonant-specific: the unprocessed condition was the best for /da, pa, ma, sa/; the target condition was the best for /ga, fa, va, za, ʒa/; and the target-conflicting condition was the best for /na, ʃa/. Perception of /ba, ta, ka/ was independent of the signal processing. The results suggest that enhancing the target ranges is an efficient way to improve consonant recognition while the removal of conflicting ranges negatively impacts consonant recognition.