Active deep learning for the identification of concepts and relations in electroencephalography reports.

The identification of medical concepts, their attributes and the relations between concepts in a large corpus of Electroencephalography (EEG) reports is a crucial step in the development of an EEG-specific patient cohort retrieval system. However, the recognition of multiple types of medical concepts, along with the many attributes characterizing them is challenging, and so is the recognition of the possible relations between them, especially when desiring to make use of active learning. To address these challenges, in this paper we present the Self-Attention Concept, Attribute and Relation (SACAR) identifier, which relies on a powerful encoding mechanism based on the recently introduced Transformer neural architecture (Dehghani et al., 2018). The SACAR identifier enabled us to consider a recently introduced framework for active learning which uses deep imitation learning for its selection policy. Our experimental results show that SACAR was able to identify medical concepts more precisely and exhibited enhanced recall, compared with previous methods. Moreover, SACAR achieves superior performance in attribute classification for attribute categories of interest, while identifying the relations between concepts with performance competitive with our previous techniques. As a multi-task network, SACAR achieves this performance on the three prediction tasks simultaneously, with a single, complex neural network. The learning curves obtained in the active learning process when using the novel Active Learning Policy Neural Network (ALPNN) show a significant increase in performance as the active learning progresses. These promising results enable the extraction of clinical knowledge available in a large collection of EEG reports.

Usefulness of the skin biopsy as a tool in the diagnosis of silvery hair syndrome.

BACKGROUND/OBJECTIVES: Silvery hair syndrome is a rare, autosomal-recessive entity characterized by silvery gray hair, eyebrows, and eyelashes and may be associated or not with immunologic or neurologic alterations. Two main types have been recognized: Chediak-Higashi syndrome and Griscelli syndrome. Hair shaft examination under light microscopy has been a useful tool to differentiate Chediak-Higashi syndrome from Griscelli syndrome, although distribution of melanin varies according to hair color related to ethnicity. The objective was to compare the pattern of melanin in the skin and with the pattern of melanin distribution in the hair shaft. METHODS: Sixteen patients with silvery hair syndrome were selected (Chediak-Higashi syndrome 5, Griscelli syndrome 11). The distribution of melanin granules in skin and hair shafts was compared and correlated with clinical diagnoses. RESULTS: Chediak-Higashi syndrome was characterized by small granules of melanin uniformly distributed throughout the thickness of the epidermis. Griscelli syndrome was characterized by an irregular pigment distribution in the epidermal basal layer with large and dense granules alternating with areas lacking melanin pigment. In two cases, study of the hair was not conclusive, but the skin showed the characteristic pattern of Griscelli syndrome. CONCLUSION: Skin biopsy is a useful tool in differentiating Chediak-Higashi syndrome from Griscelli syndrome and as a complementary study in cases in which hair shaft pigment distribution does not support the diagnosis, especially in patients with fair hair. The distribution of melanin granules in the skin correlates with that observed in the hair shaft, allowing Chediak-Higashi syndrome to be differentiated from Griscelli syndrome, at any age.

Automatic recognition of symptom severity from psychiatric evaluation records.

This paper presents a novel method for automatically recognizing symptom severity by using natural language processing of psychiatric evaluation records to extract features that are processed by machine learning techniques to assign a severity score to each record evaluated in the 2016 RDoC for Psychiatry Challenge from CEGS/N-GRID. The natural language processing techniques focused on (a) discerning the discourse information expressed in questions and answers; (b) identifying medical concepts that relate to mental disorders; and (c) accounting for the role of negation. The machine learning techniques rely on the assumptions that (1) the severity of a patient's positive valence symptoms exists on a latent continuous spectrum and (2) all the patient's answers and narratives documented in the psychological evaluation records are informed by the patient's latent severity score along this spectrum. These assumptions motivated our two-step machine learning framework for automatically recognizing psychological symptom severity. In the first step, the latent continuous severity score is inferred from each record; in the second step, the severity score is mapped to one of the four discrete severity levels used in the CEGS/N-GRID challenge. We evaluated three methods for inferring the latent severity score associated with each record: (i) pointwise ridge regression; (ii) pairwise comparison-based classification; and (iii) a hybrid approach combining pointwise regression and the pairwise classifier. The second step was implemented using a tree of cascading support vector machine (SVM) classifiers. While the official evaluation results indicate that all three methods are promising, the hybrid approach not only outperformed the pairwise and pointwise methods, but also produced the second highest performance of all submissions to the CEGS/N-GRID challenge with a normalized MAE score of 84.093% (where higher numbers indicate better performance). These evaluation results enabled us to observe that, for this task, considering pairwise information can produce more accurate severity scores than pointwise regression - an approach widely used in other systems for assigning severity scores. Moreover, our analysis indicates that using a cascading SVM tree outperforms traditional SVM classification methods for the purpose of determining discrete severity levels.

Rhabdomyomatous Mesenchymal Hamartoma: A Deep Subcutaneous Lesion in the Sternoclavicular Area.

Cutaneous rhabdomyomatous mesenchymal hamartoma (RMH) is a rare benign tumor composed of two or more types of mesenchymal-derived cells. RMHs are generally sporadic and independent, but they can be associated with congenital abnormalities. We report a subcutaneous case of RMH in the sternoclavicular area with two recurrences after complete surgical excision. The course is variable and can range from spontaneous resolution to repeated recurrences.

Hydroa vacciniforme-like lymphoma: a chronic EBV+ lymphoproliferative disorder with risk to develop a systemic lymphoma.

Hydroa vacciniforme-like lymphoma (HVLL) is an Epstein-Barr virus (EBV)-positive T-cell lymphoproliferative disorder of childhood that occurs mainly in Central and South America and Asia. We present the clinicopathological features of 20 Mexican children with HVLL with a median age of 8 years at diagnosis (range, 1-15). All patients presented with skin lesions involving sun-exposed areas, but not exclusively. Fever, lymphadenopathy, and hepatosplenomegaly were often observed. Most patients were treated with immunomodulators and/or immunosuppressive agents, resulting in temporary remission. For 13 patients follow-up was available for a median of 3 years (range, 1 month-13 years). Three patients with long follow-up (9-13 years) are alive with disease. Four patients died, 2 after developing systemic lymphoma. Histologically, the skin showed a predominantly angiocentric and periadnexal Epstein-Barr early RNA+ lymphoid infiltrate with variable atypia and subcutaneous involvement. Fifteen patients showed a T-cell phenotype (12, αß; 2, γδ; 1, silent phenotype) and monoclonal T-cell receptor-γ rearrangements, whereas 6 exhibited a natural killer (NK)-cell phenotype. Four patients had hypersensitivity to mosquito bites. One patient showed both phenotypes. HVLL is an EBV-associated lymphoproliferative disorder of αß-, γδ-, or NK-cell phenotype with a broad clinical spectrum, usually prolonged clinical course, and risk for progression to systemic disease.

Dermoid cysts: a report of 75 pediatric patients.

Dermoid cysts (DCs) are benign cutaneous tumors that tend to persist and grow. The aim of this study was to examine the clinicopathologic features of congenital DCs. We present a case series of 75 children with a clinicopathologic diagnosis of DC. Seventy-two cysts were located on the head, one on the neck, and two on the trunk. Six cysts were located along the midline. Eight patients had symptoms other than changes in cyst size. Imaging studies were performed on 15 patients. Surgical excision was the primary treatment in all 75 cases. Neurosurgery and ophthalmology services were involved in the care of some patients. Histopathologic studies reported a foreign body giant cell reaction in 17 of the cysts. No recurrence was documented. DCs can remain stable for years, but they can become symptomatic as a result of enlargement and rupture or, more rarely, as a result of extension into surrounding tissues. Physicians should be aware that certain locations have a higher risk of DC extension, and adequate diagnostic investigations should be performed before their complete resection.

Less common clinical manifestations of atopic dermatitis: prevalence by age.

The common manifestations of atopic dermatitis (AD) appear sequentially with involvement of the cheeks in infancy, flexural extremities in childhood, and hands in adulthood. Although less common clinical manifestations are well described, they have not been the subject of epidemiologic studies to describe their prevalence in specific age groups. This observational, cross-sectional, comparative study included 131 children younger than 18 of both sexes with AD who attended the clinics of the Dermatology Department of the National Institute of Pediatrics in Mexico City. Patients were examined to determine the presence of infrequent clinical manifestations of AD during infancy, preschool and school age, and adolescence and stratified according to sex, age, and number of clinical signs. A chi-square test was used to detect differences according to age and sex. Logistic regression analysis was also performed. The main findings according to age were genital dermatitis and papular-lichenoid dermatitis variant in infants; atopic feet, prurigo-like, nummular pattern, and erythroderma in preschool and school-aged children; and eyelid eczema and nipple dermatitis in adolescents. The risk of development of nipple dermatitis and eyelid eczema increased with age, and the development of genital dermatitis decreased with age. The knowledge of the prevalence of less common clinical manifestations of AD according to age in different populations might be helpful in diagnosing incipient cases of AD.

Congenital cutaneous angioleiomyoma.

Congenital cutaneous angioleiomyoma is an extremely rare benign smooth muscle tumor. We present a case of a firm, painful subcutaneous mass noticed at birth on the left leg that on surgical excision proved to be an angioleiomyoma. Prognosis is good, and recurrences are uncommon. To our knowledge, this is the second report of a congenital angioleiomyoma.

The impact of learning Unified Medical Language System knowledge embeddings in relation extraction from biomedical texts.

OBJECTIVE: We explored how knowledge embeddings (KEs) learned from the Unified Medical Language System (UMLS) Metathesaurus impact the quality of relation extraction on 2 diverse sets of biomedical texts. MATERIALS AND METHODS: Two forms of KEs were learned for concepts and relation types from the UMLS Metathesaurus, namely lexicalized knowledge embeddings (LKEs) and unlexicalized KEs. A knowledge embedding encoder (KEE) enabled learning either LKEs or unlexicalized KEs as well as neural models capable of producing LKEs for mentions of biomedical concepts in texts and relation types that are not encoded in the UMLS Metathesaurus. This allowed us to design the relation extraction with knowledge embeddings (REKE) system, which incorporates either LKEs or unlexicalized KEs produced for relation types of interest and their arguments. RESULTS: The incorporation of either LKEs or unlexicalized KE in REKE advances the state of the art in relation extraction on 2 relation extraction datasets: the 2010 i2b2/VA dataset and the 2013 Drug-Drug Interaction Extraction Challenge corpus. Moreover, the impact of LKEs is superior, achieving F1 scores of 78.2 and 82.0, respectively. DISCUSSION: REKE not only highlights the importance of incorporating knowledge encoded in the UMLS Metathesaurus in a novel way, through 2 possible forms of KEs, but it also showcases the subtleties of incorporating KEs in relation extraction systems. CONCLUSIONS: Incorporating LKEs informed by the UMLS Metathesaurus in a relation extraction system operating on biomedical texts shows significant promise. We present the REKE system, which establishes new state-of-the-art results for relation extraction on 2 datasets when using LKEs.

Adversarial Learning of Knowledge Embeddings for the Unified Medical Language System.

Incorporating the knowledge encoded in the Unified Medical Language System (UMLS) in deep learning methods requires learning knowledge embeddings from the knowledge graphs available in UMLS: the Metathesaurus and the Semantic Network. In this paper we present a technique using Generative Adversarial Networks (GANs) for learning UMLS embeddings and showcase their usage in a clinical prediction model. When the UMLS embeddings are available, the predictions improve by up to 6.9% absolute F1 score.

Bootstrapping Adversarial Learning of Biomedical Ontology Alignments.

Learning how to automatically align biomedical ontologies has been a long-standing goal, given their ever-growing content and the many applications that rely on them. Because the knowledge graphs underlying biomedical ontologies enable neural learning techniques to acquire knowledge embeddings as representations of these ontologies, neural learning can also consider ontology alignments. In this paper, we present the Knowledge-graph Alignment & Embedding Generative Adversarial Network (KAEGAN) which learns (a) to represent the relational knowledge from two distinct biomedical ontologies in the form of knowledge embeddings and (b) to use them for ontology alignment, by also relying on the ontology semantics. KAEGAN is a Generative Adversarial Network trained using bootstrapping to iteratively improve the learned alignments. Experimental results show promise, demonstrating that jointly learning ontology alignment and knowledge representation improves upon learning either in isolation.

Memory-Augmented Active Deep Learning for Identifying Relations Between Distant Medical Concepts in Electroencephalography Reports.

The automatic identification of relations between medical concepts in a large corpus of Electroencephalography (EEG) reports is an important step in the development of an EEG-specific patient cohort retrieval system as well as in the acquisition of EEG-specific knowledge from this corpus. EEG-specific relations involve medical concepts that are not typically mentioned in the same sentence or even the same section of a report, thus requiring extraction techniques that can handle such long-distance dependencies. To address this challenge, we present a novel frame work which combines the advantages of a deep learning framework employing Dynamic Relational Memory (DRM) with active learning. While DRM enables the prediction of long-distance relations, active learning provides a mechanism for accurately identifying relations with minimal training data, obtaining an 5-fold cross validationF1 score of 0.7475 on a set of 140 EEG reports selected with active learning. The results obtained with our novel framework show great promise.

Keratosis lichenoides chronica in pediatric patients: a different disease?

Keratosis lichenoides chronica (KLC) is a rare acquired disease of adulthood, of unknown etiology, characterized by keratotic parallel linear lesions, retiform plaques, and keratotic, often follicular papules, chronicity and lichenoid histopathologic features. KLC of pediatric onset is considered extremely rare. Its features and relationship to adult onset KLC are unknown. We studied 8 cases of pediatric-onset KLC in the literature and 6 personal cases and compared them with 40 reported adult-onset KLC patients. The following features characterize pediatric-onset KLC: familial occurrence; probable autosomal recessive inheritance; early or congenital onset with facial erythemato-purpuric macules; forehead, eyebrow, and eyelash alopecia; pruritus; and a low frequency of other cutaneous and systemic abnormalities. Pediatric-onset KLC may represent a different disease or a subset of adult-onset KLC, with special genetic and clinical characteristics. Determining its precise nosology will have prognostic and therapeutic implications.

Hypomelanotic conditions of the newborn and infant.

Depigmented nevi, pityriasis alba, and postinflammatory hypopigmentation are the most frequent hypomelanotic conditions in newborns and infants. These, and examples of less frequent hypopigmentations are briefly discussed in this article. A new classification for depigmented nevi is also proposed.

Active Deep Learning-Based Annotation of Electroencephalography Reports for Cohort Identification.

The annotation of a large corpus of Electroencephalography (EEG) reports is a crucial step in the development of an EEG-specific patient cohort retrieval system. The annotation of multiple types of EEG-specific medical concepts, along with their polarity and modality, is challenging, especially when automatically performed on Big Data. To address this challenge, we present a novel framework which combines the advantages of active and deep learning while producing annotations that capture a variety of attributes of medical concepts. Results obtained through our novel framework show great promise.

Deep Learning Meets Biomedical Ontologies: Knowledge Embeddings for Epilepsy.

While biomedical ontologies have traditionally been used to guide the identification of concepts or relations in biomedical data, recent advances in deep learning are able to capture high-quality knowledge from textual data and represent it in graphical structures. As opposed to the top-down methodology used in the generation of ontologies, which starts with the principled design of the upper ontology, the bottom-up methodology enabled by deep learning encodes the likelihood that concepts share certain relations, as evidenced by data. In this paper, we present a knowledge representation produced by deep learning methods, called Medical Knowledge Embeddings (MKE), that encode medical concepts related to the study of epilepsy and the relations between them. Many of the epilepsy-relevant medical concepts from MKE are not yet available in existing biomedical ontologies, but are mentioned in vast collections of epilepsy-related medical records which also imply their relationships. The evaluation of the MKE indicates high accuracy of the medical concepts automatically identified from clinical text as well as promising results in terms of correctness and completeness of relations produced by deep learning.

Necrotizing fasciitis: report of 39 pediatric cases.

BACKGROUND: Necrotizing fasciitis (NF) is a severe, life-threatening soft tissue infection. General features and risk factors for fatal outcome in children are not well known. OBJECTIVE: To characterize the features of NF in children and the risk factors for fatal outcome. DESIGN: Retrospective, comparative, observational, and longitudinal trial. SETTING: Dermatology department of a tertiary care pediatric hospital. PATIENTS: All patients with clinical and/or histopathological diagnosis of NF seen from January 1, 1971, through December 31, 2000. MAIN OUTCOME VARIABLES: Incidence, age, sex, number and location of lesions, preexisting conditions, initiating factors, clinical and laboratory features, diagnosis at admission, treatment, evolution, sequelae, and risk factors for fatal outcome. RESULTS: We examined 39 patients with NF (0.018% of all hospitalized patients). Twenty-one patients (54%) were boys. Mean age was 4.4 years. Single lesions were seen in 30 (77%) of patients, with 21(54%) in extremities. The most frequent preexisting condition was malnutrition in 14 patients (36%). The most frequent initiating factor was varicella in 13 patients (33%). Diagnosis of NF at admission was made in 11 patients (28%). Bacterial isolations in 24 patients (62%) were polymicrobial in 17 (71%). Pseudomonas aeruginosa was the most frequently isolated bacteria; gram-negative isolates, the most frequently associated bacteria. Complications were present in 33 patients (85%), mortality in 7 (18%), and sequelae in 29 (91%) of 32 surviving patients. The significant risk factor related to a fatal outcome was immunosuppression. CONCLUSIONS: Necrotizing fasciitis in children is frequently misdiagnosed, and several features differ from those of NF in adults. Immunosuppression was the main factor related to death. Early surgical debridement and antibiotics were the most important therapeutic measures.

Medallion-like dermal dendrocyte hamartoma: a new clinically and histopathologically distinct lesion.

Dermal dendrocyte hamartomas are extremely rare; only two examples have been described with clinical features different from our cases and with incomplete immunohistochemical characterization. We report three female patients presenting a medallion-shaped, well-defined, slightly atrophic and asymptomatic congenital lesion. All 3 patients showed a fusiform-cell proliferation. Immunohistochemistry was positive for CD34, factor XIIIa, and fascin. Electron microscopy showed typical features of dermal dendrocytes. We believe that the lesions described represent a new, clinically and histopathologically distinct lesion originating in dermal dendrocytes. We propose to name it medallion-like dermal dendrocyte hamartoma.

Phakomatosis pigmentovascularis II A and II B: clinical findings in 24 patients.

Nearly 200 cases of phakomatosis pigmentovascularis (PPV) have been reported worldwide, most of them of Japanese origin. There are 5 types and 10 subtypes of PPV. Its etiology might be explained by the twin spotting phenomenon. The relative frequency of PPV at the National Institute of Pediatrics was 5.8 per 100,000 pediatric patients and 0.634 per 100,000 dermatological patients. We report 24 cases of PPV with an average follow up of 5 years and the following findings: PPV type II A in 4 male and 2 female patients with melanosis bulbi in 3 and glaucoma in 1. PPV type II B in 7 male and 11 female patients, with melanosis bulbi in 9, glaucoma in 9, iris mammillations in 2, Sturge Weber syndrome in 6 female patients, and Klippel-Trenaunay syndrome in 2 males, hemifacial, hemicorporal, or limb hypertrophy without venous insufficiency in 6 female and 4 male patients. During the follow-up time of 60 months, progressive fading of melanotic and vascular macules were observed in 7 patients. No other types of PPV were found. Systemic involvement in PPV was related to the body surface area affected by the vascular macules. Ectodermal and mesodermal migration disorders might be involved in the pathogenesis of PPV.

Oral desensitization in papular urticaria in children.

Papular urticaria (PU) is among the commonest skin ailments in children. Induced specific desensitization to insect bites is theoretically an effective means of prevention of PU. In this double blind placebo controlled study, an oral vaccine prepared from insect saliva was compared with placebo (stable vaccine solvent). Vaccine and placebo effectiveness were tested by counting active PU lesions, serum eosinophils, and IgE, before and after 4 months of treatment. Statistically significant differences between oral vaccine and placebo were not found in the clinical or the immunological variables tested. We conclude that, although a lack of oral vaccine efficacy was suspected, larger study samples are needed to strengthen our conclusion.