Deep Learning for Automated Image Segmentation of the Middle Ear: A Scoping Review.

OBJECTIVE: Convolutional neural networks (CNNs) have revolutionized medical image segmentation in recent years. This scoping review aimed to carry out a comprehensive review of the literature describing automated image segmentation of the middle ear using CNNs from computed tomography (CT) scans. DATA SOURCES: A comprehensive literature search, generated jointly with a medical librarian, was performed on Medline, Embase, Scopus, Web of Science, and Cochrane, using Medical Subject Heading terms and keywords. Databases were searched from inception to July 2023. Reference lists of included papers were also screened. REVIEW METHODS: Ten studies were included for analysis, which contained a total of 866 scans which were used in model training/testing. Thirteen different architectures were described to perform automated segmentation. The best Dice similarity coefficient (DSC) for the entire ossicular chain was 0.87 using ResNet. The highest DSC for any structure was the incus using 3D-V-Net at 0.93. The most difficult structure to segment was the stapes, with the highest DSC of 0.84 using 3D-V-Net. CONCLUSIONS: Numerous architectures have demonstrated good performance in segmenting the middle ear using CNNs. To overcome some of the difficulties in segmenting the stapes, we recommend the development of an architecture trained on cone beam CTs to provide improved spatial resolution to assist with delineating the smallest ossicle. IMPLICATIONS FOR PRACTICE: This has clinical applications for preoperative planning, diagnosis, and simulation.

Named entity recognition of pharmacokinetic parameters in the scientific literature.

The development of accurate predictions for a new drug's absorption, distribution, metabolism, and excretion profiles in the early stages of drug development is crucial due to high candidate failure rates. The absence of comprehensive, standardised, and updated pharmacokinetic (PK) repositories limits pre-clinical predictions and often requires searching through the scientific literature for PK parameter estimates from similar compounds. While text mining offers promising advancements in automatic PK parameter extraction, accurate Named Entity Recognition (NER) of PK terms remains a bottleneck due to limited resources. This work addresses this gap by introducing novel corpora and language models specifically designed for effective NER of PK parameters. Leveraging active learning approaches, we developed an annotated corpus containing over 4000 entity mentions found across the PK literature on PubMed. To identify the most effective model for PK NER, we fine-tuned and evaluated different NER architectures on our corpus. Fine-tuning BioBERT exhibited the best results, achieving a strict F 1 score of 90.37% in recognising PK parameter mentions, significantly outperforming heuristic approaches and models trained on existing corpora. To accelerate the development of end-to-end PK information extraction pipelines and improve pre-clinical PK predictions, the PK NER models and the labelled corpus were released open source at https://github.com/PKPDAI/PKNER .

Delineating COVID-19 subgroups using routine clinical data identifies distinct in-hospital outcomes.

The COVID-19 pandemic has been a great challenge to healthcare systems worldwide. It highlighted the need for robust predictive models which can be readily deployed to uncover heterogeneities in disease course, aid decision-making and prioritise treatment. We adapted an unsupervised data-driven model-SuStaIn, to be utilised for short-term infectious disease like COVID-19, based on 11 commonly recorded clinical measures. We used 1344 patients from the National COVID-19 Chest Imaging Database (NCCID), hospitalised for RT-PCR confirmed COVID-19 disease, splitting them equally into a training and an independent validation cohort. We discovered three COVID-19 subtypes (General Haemodynamic, Renal and Immunological) and introduced disease severity stages, both of which were predictive of distinct risks of in-hospital mortality or escalation of treatment, when analysed using Cox Proportional Hazards models. A low-risk Normal-appearing subtype was also discovered. The model and our full pipeline are available online and can be adapted for future outbreaks of COVID-19 or other infectious disease.

Frequency tuning of medial-olivocochlear-efferent acoustic reflexes in humans as functions of probe frequency.

The medial-olivocochlear (MOC) acoustic reflex is thought to provide frequency-specific feedback that adjusts the gain of cochlear amplification, but little is known about how frequency specific the reflex actually is. We measured human MOC tuning through changes in stimulus frequency otoacoustic emissions (SFOAEs) from 40-dB-SPL tones at probe frequencies (f(p)s) near 0.5, 1.0, and 4.0 kHz. MOC activity was elicited by 60-dB-SPL ipsilateral, contralateral, or bilateral tones or half-octave noise bands, with elicitor frequency (f(e)) varied in half-octave steps. Tone and noise elicitors produced similar results. At all probe frequencies, SFOAE changes were produced by a wide range of elicitor frequencies with elicitor frequencies near 0.7-2.0 kHz being particularly effective. MOC-induced changes in SFOAE magnitude and SFOAE phase were surprisingly different functions of f(e): magnitude inhibition largest for f(e) close to f(p), phase change largest for f(e) remote from f(p). The metric ΔSFOAE, which combines both magnitude and phase changes, provided the best match to reported (cat) MOC neural inhibition. Ipsilateral and contralateral MOC reflexes often showed dramatic differences in plots of MOC effect vs. elicitor frequency, indicating that the contralateral reflex does not give an accurate picture of ipsilateral-reflex properties. These differences in MOC effects appear to imply that ipsilateral and contralateral reflexes have different actions in the cochlea. The implication of these results for MOC function, cochlear mechanics, and the production of SFOAEs are discussed.

Clinical deployment environments: Five pillars of translational machine learning for health.

Machine Learning for Health (ML4H) has demonstrated efficacy in computer imaging and other self-contained digital workflows, but has failed to substantially impact routine clinical care. This is no longer because of poor adoption of Electronic Health Records Systems (EHRS), but because ML4H needs an infrastructure for development, deployment and evaluation within the healthcare institution. In this paper, we propose a design pattern called a Clinical Deployment Environment (CDE). We sketch the five pillars of the CDE: (1) real world development supported by live data where ML4H teams can iteratively build and test at the bedside (2) an ML-Ops platform that brings the rigour and standards of continuous deployment to ML4H (3) design and supervision by those with expertise in AI safety (4) the methods of implementation science that enable the algorithmic insights to influence the behaviour of clinicians and patients and (5) continuous evaluation that uses randomisation to avoid bias but in an agile manner. The CDE is intended to answer the same requirements that bio-medicine articulated in establishing the translational medicine domain. It envisions a transition from "real-world" data to "real-world" development.

Reduced oxygen saturation entropy is associated with poor prognosis in critically ill patients with sepsis.

Recent studies have found that oxygen saturation (SpO2 ) variability analysis has potential for noninvasive assessment of the functional connectivity of cardiorespiratory control systems during hypoxia. Patients with sepsis have suboptimal tissue oxygenation and impaired organ system connectivity. Our objective with this report was to highlight the potential use for SpO2 variability analysis in predicting intensive care survival in patients with sepsis. MIMIC-III clinical data of 164 adults meeting Sepsis-3 criteria and with 30 min of SpO2 and respiratory rate data were analyzed. The complexity of SpO2 signals was measured through various entropy calculations such as sample entropy and multiscale entropy analysis. The sequential organ failure assessment (SOFA) score was calculated to assess the severity of sepsis and multiorgan failure. While the standard deviation of SpO2 signals was comparable in the non-survivor and survivor groups, non-survivors had significantly lower SpO2 entropy than those who survived their ICU stay (0.107 ± 0.084 vs. 0.070 ± 0.083, p < 0.05). According to Cox regression analysis, higher SpO2 entropy was a predictor of survival in patients with sepsis. Multivariate analysis also showed that the prognostic value of SpO2 entropy was independent of other covariates such as age, SOFA score, mean SpO2 , and ventilation status. When SpO2 entropy was combined with mean SpO2 , the composite had a significantly higher performance in prediction of survival. Analysis of SpO2 entropy can predict patient outcome, and when combined with SpO2 mean, can provide improved prognostic information. The prognostic power is on par with the SOFA score. This analysis can easily be incorporated into current ICU practice and has potential for noninvasive assessment of critically ill patients.

Phenotyping of ABCA4 Retinopathy by Machine Learning Analysis of Full-Field Electroretinography.

Purpose: Biallelic pathogenic variants in ABCA4 are the commonest cause of monogenic retinal disease. The full-field electroretinogram (ERG) quantifies severity of retinal dysfunction. We explored application of machine learning in ERG interpretation and in genotype-phenotype correlations. Methods: International standard ERGs in 597 cases of ABCA4 retinopathy were classified into three functional phenotypes by human experts: macular dysfunction alone (group 1), or with additional generalized cone dysfunction (group 2), or both cone and rod dysfunction (group 3). Algorithms were developed for automatic selection and measurement of ERG components and for classification of ERG phenotype. Elastic-net regression was used to quantify severity of specific ABCA4 variants based on effect on retinal function. Results: Of the cohort, 57.6%, 7.4%, and 35.0% fell into groups 1, 2, and 3 respectively. Compared with human experts, automated classification showed overall accuracy of 91.8% (SE, 0.169), and 96.7%, 39.3%, and 93.8% for groups 1, 2, and 3. When groups 2 and 3 were combined, the average holdout group accuracy was 93.6% (SE, 0.142). A regression model yielded phenotypic severity scores for the 47 commonest ABCA4 variants. Conclusions: This study quantifies prevalence of phenotypic groups based on retinal function in a uniquely large single-center cohort of patients with electrophysiologically characterized ABCA4 retinopathy and shows applicability of machine learning. Novel regression-based analyses of ABCA4 variant severity could identify individuals predisposed to severe disease. Translational Relevance: Machine learning can yield meaningful classifications of ERG data, and data-driven scoring of genetic variants can identify patients likely to benefit most from future therapies.

An automated approach to identify scientific publications reporting pharmacokinetic parameters.

Pharmacokinetic (PK) predictions of new chemical entities are aided by prior knowledge from other compounds. The development of robust algorithms that improve preclinical and clinical phases of drug development remains constrained by the need to search, curate and standardise PK information across the constantly-growing scientific literature. The lack of centralised, up-to-date and comprehensive repositories of PK data represents a significant limitation in the drug development pipeline.In this work, we propose a machine learning approach to automatically identify and characterise scientific publications reporting PK parameters from in vivo data, providing a centralised repository of PK literature. A dataset of 4,792 PubMed publications was labelled by field experts depending on whether in vivo PK parameters were estimated in the study. Different classification pipelines were compared using a bootstrap approach and the best-performing architecture was used to develop a comprehensive and automatically-updated repository of PK publications. The best-performing architecture encoded documents using unigram features and mean pooling of BioBERT embeddings obtaining an F1 score of 83.8% on the test set. The pipeline retrieved over 121K PubMed publications in which in vivo PK parameters were estimated and it was scheduled to perform weekly updates on newly published articles. All the relevant documents were released through a publicly available web interface (https://app.pkpdai.com) and characterised by the drugs, species and conditions mentioned in the abstract, to facilitate the subsequent search of relevant PK data. This automated, open-access repository can be used to accelerate the search and comparison of PK results, curate ADME datasets, and facilitate subsequent text mining tasks in the PK domain.

Medial olivocochlear efferent reflex in humans: otoacoustic emission (OAE) measurement issues and the advantages of stimulus frequency OAEs.

Otoacoustic emissions (OAEs) are useful for studying medial olivocochlear (MOC) efferents, but several unresolved methodological issues cloud the interpretation of the data they produce. Most efferent assays use a "probe stimulus" to produce an OAE and an "elicitor stimulus" to evoke efferent activity and thereby change the OAE. However, little attention has been given to whether the probe stimulus itself elicits efferent activity. In addition, most studies use only contralateral ( re the probe) elicitors and do not include measurements to rule out middle-ear muscle (MEM) contractions. Here we describe methods to deal with these problems and present a new efferent assay based on stimulus frequency OAEs (SFOAEs) that incorporates these methods. By using a postelicitor window, we make measurements in individual subjects of efferent effects from contralateral, ipsilateral, and bilateral elicitors. Using our SFOAE assay, we demonstrate that commonly used probe sounds (clicks, tone pips, and tone pairs) elicit efferent activity, by themselves. Thus, results of efferent assays using these probe stimuli can be confounded by unwanted efferent activation. In contrast, the single 40 dB SPL tone used as the probe sound for SFOAE-based measurements evoked little or no efferent activity. Since they evoke efferent activation, clicks, tone pips, and tone pairs can be used in an adaptation efferent assay, but such paradigms are limited in measurement scope compared to paradigms that separate probe and elicitor stimuli. Finally, we describe tests to distinguish middle-ear muscle (MEM) effects from MOC effects for a number of OAE assays and show results from SFOAE-based tests. The SFOAE assay used in this study provides a sensitive, flexible, frequency-specific assay of medial efferent activation that uses a low-level probe sound that elicits little or no efferent activity, and thus provides results that can be interpreted without the confound of unintended efferent activation.

Reflex control of the human inner ear: a half-octave offset in medial efferent feedback that is consistent with an efferent role in the control of masking.

The high sensitivity and frequency selectivity of the mammalian cochlea is due to amplification produced by outer hair cells (OHCs) and controlled by medial olivocochlear (MOC) efferents. Data from animals led to the view that MOC fibers provide frequency-specific inhibitory feedback; however, these studies did not measure intact MOC reflexes. To test whether MOC inhibition is primarily at the frequency that elicits the MOC activity, acoustically elicited MOC effects were quantified in humans by the change in otoacoustic emissions produced by 60-dB SPL tone and half-octave-band noise elicitors at different frequencies relative to a 40-dB SPL, 1-kHz probe tone. On average, all elicitors produced MOC effects that were skewed (elicitor frequencies -1 octave below the probe produced larger effects than those -1 octave above). The largest MOC effects were from elicitors below the probe frequency for contra- and bilateral elicitors but were from elicitors centered at the probe frequency for ipsilateral elicitors. Typically, ipsilateral elicitors produced larger effects than contralateral elicitors and bilateral elicitors produced effects near the ipsi+contra sum. Elicitors at levels down to 30-dB SPL produced similar patterns. Tuning curves (TCs) interpolated from these data were V-shaped with Q10s approximately 2. These are sharper than MOC-fiber TCs found near 1 kHz in cats and guinea pigs. Because cochlear amplification is skewed (more below the best frequency of a cochlear region), these data are consistent with an anti-masking role of MOC efferents that reduces masking by reducing the cochlear amplification seen at 1 kHz.

Human medial olivocochlear reflex: effects as functions of contralateral, ipsilateral, and bilateral elicitor bandwidths.

Animal studies have led to the view that the acoustic medial olivocochlear (MOC) efferent reflex provides sharply tuned frequency-specific feedback that inhibits cochlear amplification. To determine if MOC activation is indeed narrow band, we measured the MOC effects in humans elicited by 60-dB sound pressure level (SPL) contralateral, ipsilateral, and bilateral noise bands as a function of noise bandwidth from 1/2 to 6.7 octaves. MOC effects were quantified by the change in stimulus frequency otoacoustic emissions from 40 dB SPL probe tones near 0.5, 1, and 4 kHz. In a second experiment, the noise bands were centered 2 octaves below probe frequencies near 1 and 4 kHz. In all cases, the MOC effects increased as elicitor bandwidth increased, with the effect saturating at about 4 octaves. Generally, the MOC effects increased as the probe frequency decreased, opposite expectations based on MOC innervation density in the cochlea. Bilateral-elicitor effects were always the largest. The ratio of ipsilateral/contralateral effects depended on elicitor bandwidth; the ratio was large for narrow-band probe-centered elicitors and approximately unity for wide-band elicitors. In another experiment, the MOC effects from increasing elicitor bandwidths were calculated from measurements of the MOC effects from adjacent half-octave noise bands. The predicted bandwidth function agreed well with the measured bandwidth function for contralateral elicitors, but overestimated it for ipsilateral and bilateral elicitors. Overall, the results indicate that (1) the MOC reflexes integrate excitation from almost the entire cochlear length, (2) as elicitor bandwidth is increased, the excitation from newly stimulated cochlear regions more than overcomes the reduced excitation at frequencies in the center of the elicitor band, and (3) contralateral, ipsilateral, and bilateral elicitors show MOC reflex spatial summation over most of the cochlea, but ipsilateral spatial summation is less additive than contralateral.