Search

Artificial intelligence (AI)‐based diagnostic algorithms have achieved ambitious aims through automated image pattern recognition. For neurological disorders, this includes neurodegeneration and inflammation. Scalable imaging technology for big data in neurology is optical coherence tomography (OCT). We highlight that OCT changes observed in the retina, as a window to the brain, are small, requiring rigorous quality control pipelines. There are existing tools for this purpose. Firstly, there are human‐led validated consensus quality control criteria (OSCAR‐IB) for OCT. Secondly, these criteria are embedded into OCT reporting guidelines (APOSTEL). The use of the described annotation of failed OCT scans advances machine learning. This is illustrated through the present review of the advantages and disadvantages of AI‐based applications to OCT data. The neurological conditions reviewed here for the use of big data include Alzheimer disease, stroke, multiple sclerosis (MS), Parkinson disease, and epilepsy. It is noted that while big data is relevant for AI, ownership is complex. For this reason, we also reached out to involve representatives from patient organizations and the public domain in addition to clinical and research centers. The evidence reviewed can be grouped in a five‐point expansion of the OSCAR‐IB criteria to embrace AI (OSCAR‐AI). The review concludes by specific recommendations on how this can be achieved practically and in compliance with existing guidelines.

Background Rare Eye Diseases (RED) are the leading cause of visual impairment and blindness for children and young adults in Europe. This heterogeneous group of conditions includes over 900 disorders ranging from relatively prevalent disorders such as retinitis pigmentosa to very rare entities such as developmental eye anomalies. A significant number of patients with RED have an underlying genetic etiology. One of the aims of the European Reference Network for Rare Eye Diseases (ERN–EYE) is to facilitate improvement in diagnosis of RED in European member states. Main body Technological advances have allowed genetic and genomic testing for RED. The outcome of genetic testing allows better understanding of the condition and allows reproductive and therapeutic options. The increase of the number of clinical trials for RED has provided urgency for genetic testing in RED. A survey of countries participating in ERN-EYE demonstrated that the majority are able to access some forms of genomic testing. However, there is significant variability, particularly regarding testing as part of clinical service. Some countries have a well-delineated rare disease pathway and have a national plan for rare diseases combined or not with a national plan for genomics in medicine. In other countries, there is a well-established organization of genetic centres that offer reimbursed genomic testing of RED and other rare diseases. Clinicians often rely upon research-funded laboratories or private companies. Notably, some member states rely on cross-border testing by way of an academic research project. Consequently, many clinicians are either unable to access testing or are confronted with long turnaround times. Overall, while the cost of sequencing has dropped, the cumulative cost of a genomic testing service for populations remains considerable. Importantly, the majority of countries reported healthcare budgets that limit testing. Short conclusion Despite technological advances, critical gaps in genomic testing remain in Europe, especially in smaller countries where no formal genomic testing pathways exist. Even within larger countries, the existing arrangements are insufficient to meet the demand and to ensure access. ERN-EYE promotes access to genetic testing in RED and emphasizes the clinical need and relevance of genetic testing in RED.

This article illustrates the cross-modal application of the phonotactic no geminate constraint that disallows geminate consonants within a prosodic word in German. In phonological production, forms like |hat+tə| ‘have-3SG.PST’ surface as /hatə/ (instead of */hatːə/) due to this constraint. In speech perception, phonetic forms with geminates like [bʁoːtːaɪk] ‘bread dough’ are perceived as consisting of two prosodic words because of this constraint. And in the reading process, orthographic forms with double consonantal graphemes like and are read as /wal/ and /tɛlɐ/, not */walː/ and */tɛlːɐ/ due to this constraint. I show that this observation can be formalized in Optimality Theory by applying the same phonotactic constraint in phonological production, in the perception grammar (Boersma 2007) and in the reading grammar (Hamann & Colombo 2017). In both the perception and the reading grammar, this constraint interacts with constraints that handle the arbitrary mapping between the sensory input (auditory or written) and a surface phonological form. This approach is shown to be preferable to previous analyses of reading double consonantal graphemes in German, which either lack an explicit formalization of the phonological knowledge or reduplicate this in their phoneme-to-grapheme mappings.

The pre-NC vowel in many Bantu languages is generally understood to be long. In Bemba, where there is also a vowel length contrast, this raises the question whether the pre-NC vowel is phonetically as long as lexical long vowels and how phonologised this length might be. In contrast to lexically long vowels, pre-NC vowel length is attributed to vowel lengthening resulting from the restructuring of a nasal to create a prenasalised stop. This is thus relevant to whether such stops are treated as unit segments or not. The present paper focuses on the pre-NC vowel and presents an evaluation of whether the pre-NC vowel is monomoraic or bimoraic by considering both phonetic and phonological evidence. Segmental and most tonal evidence leans towards a monomoraic treatment of the pre-NC vowel. One set of tonal data, however, shows variation in moraicity, presenting a mixed picture that we conclude emanates from the intermediate phonetic duration of the pre-NC vowel.

BackgroundNonophthalmologist physicians do not confidently perform direct ophthalmoscopy. The use of artificial intelligence to detect papilledema and other optic-disk abnormalities from fundus photographs has not been well studied.MethodsWe trained, validated, and externally tested a deep-learning system to classify optic disks as being normal or having papilledema or other abnormalities from 15,846 retrospectively collected ocular fundus photographs that had been obtained with pharmacologic pupillary dilation and various digital cameras in persons from multiple ethnic populations. Of these photographs, 14,341 from 19 sites in 11 countries were used for training and validation, and 1505 photographs from 5 other sites were used for external testing. Performance at classifying the optic-disk appearance was evaluated by calculating the area under the receiver-operating-characteristic curve (AUC), sensitivity, and specificity, as compared with a reference standard of clinical diagnoses by neuro-ophthalmologists.ResultsThe training and validation data sets from 6779 patients included 14,341 photographs: 9156 of normal disks, 2148 of disks with papilledema, and 3037 of disks with other abnormalities. The percentage classified as being normal ranged across sites from 9.8 to 100%; the percentage classified as having papilledema ranged across sites from zero to 59.5%. In the validation set, the system discriminated disks with papilledema from normal disks and disks with nonpapilledema abnormalities with an AUC of 0.99 (95% confidence interval [CI], 0.98 to 0.99) and normal from abnormal disks with an AUC of 0.99 (95% CI, 0.99 to 0.99). In the external-testing data set of 1505 photographs, the system had an AUC for the detection of papilledema of 0.96 (95% CI, 0.95 to 0.97), a sensitivity of 96.4% (95% CI, 93.9 to 98.3), and a specificity of 84.7% (95% CI, 82.3 to 87.1).ConclusionsA deep-learning system using fundus photographs with pharmacologically dilated pupils differentiated among optic disks with papilledema, normal disks, and disks with nonpapilledema abnormalities. (Funded by the Singapore National Medical Research Council and the SingHealth Duke–NUS Ophthalmology and Visual Sciences Academic Clinical Program.)

Background The optical accessibility of the eye and technological advances in ophthalmic diagnostics have put ophthalmology at the forefront of data-driven medicine. The focus of this study is rare eye disorders, a group of conditions whose clinical heterogeneity and geographic dispersion make data-driven, evidence-based practice particularly challenging. Inter-institutional collaboration and information sharing is crucial but the lack of standardised terminology poses an important barrier. Ontologies are computational tools that include sets of vocabulary terms arranged in hierarchical structures. They can be used to provide robust terminology standards and to enhance data interoperability. Here, we discuss the development of the ophthalmology-related component of two well-established biomedical ontologies, the Human Phenotype Ontology (HPO; includes signs, symptoms and investigation findings) and the Orphanet Rare Disease Ontology (ORDO; includes rare disease nomenclature/nosology). Methods A variety of approaches were used including automated matching to existing resources and extensive manual curation. To achieve the latter, a study group including clinicians, patient representatives and ontology developers from 17 countries was formed. A broad range of terms was discussed and validated during a dedicated workshop attended by 60 members of the group. Results A comprehensive, structured and well-defined set of terms has been agreed on including 1106 terms relating to ocular phenotypes (HPO) and 1202 terms relating to rare eye disease nomenclature (ORDO). These terms and their relevant annotations can be accessed in http://www.human-phenotype-ontology.org/ and http://www.orpha.net/; comments, corrections, suggestions and requests for new terms can be made through these websites. This is an ongoing, community-driven endeavour and both HPO and ORDO are regularly updated. Conclusions To our knowledge, this is the first effort of such scale to provide terminology standards for the rare eye disease community. We hope that this work will not only improve coding and standardise information exchange in clinical care and research, but also it will catalyse the transition to an evidence-based precision ophthalmology paradigm. Electronic supplementary material The online version of this article (10.1186/s13023-018-0980-6) contains supplementary material, which is available to authorized users.

We present a neural network model of phonetic and phonological acquisition that can handle two distinct phenomena: category creation and auditory dispersion. Within a single neural network, learning proceeds in two stages. The first stage is distributional learning, during which the model induces phonological features from an auditory input distribution; in the second stage, the model acquires knowledge about the relation between lexical categories and the auditory input distribution. The model can be used bidirectionally: once perceptual learning is complete, the network can also be asked to speak. In the production direction, effortful, perceptually peripheral tokens are avoided. In a chain of iterated learners, in which the output of one generation serves as the input to the next, sound systems emerge that maintain sufficient contrast at a moderate articulatory cost, regardless of the initial distribution.

The Bantu language Tumbuka, spoken in Malawi and Zambia, has the vowel phonemes /i, e, a, o, u/ but no phonemic contrast in vowel length. In the present study we measured the duration of vowels before homorganic nasal-obstruent sequences (NC) and compared them to the duration of vowels before single obstruents for four speakers of Tumbuka. With this data, we tested whether there is acoustic support for the existence of a phonological process of pre-NC lengthening, as reported for many other Bantu languages. Our results provide no support for such an interpretation in Tumbuka: pre-NC vowels were only 10 ms longer than pre-obstruent vowels, and though this difference was statistically significant, it is below the just noticeable difference for duration of 25 ms and therefore most likely not perceivable. We conclude that the observed pre-NC lengthening in Tumbuka is a purely co-articulatory, phonetic process.

Italian native speakers often produce short, schwalike vowels at the end of consonant-final loanwords from English. While past studies on these paragogic vowels investigated intra-speaker variation and looked at factors such as stress, voicing or intonation contours, the present study focuses on variation between speakers. We test the hypothesis that the amount of received native English input predicts how often Italians produce paragogic vowels in English loanwords, with less input causing more inserted vowels. L2 input was estimated on the basis of a questionnaire on the active use and passive exposure to English. Twenty-one Italian native speakers with varying levels of English filled out this questionnaire and took part in an irregular-plural elicitation task containing consonant-final loanwords from English. Our results show that Italian speakers with a higher self-reported level of English exposure produce fewer paragogic vowels, thereby confirming our hypothesis.