What is the state of the art of computer vision-assisted cytology? A Systematic Literature Review.

Cytology is a low-cost and non-invasive diagnostic procedure employed to support the diagnosis of a broad range of pathologies. Cells are harvested from tissues by aspiration or scraping, and it is still predominantly performed manually by medical or laboratory professionals extensively trained for this purpose. It is a time-consuming and repetitive process where many diagnostic criteria are subjective and vulnerable to human interpretation. Computer Vision technologies, by automatically generating quantitative and objective descriptions of examinations' contents, can help minimize the chances of misdiagnoses and shorten the time required for analysis. To identify the state-of-art of computer vision techniques currently applied to cytology, we conducted a Systematic Literature Review, searching for approaches for the segmentation, detection, quantification, and classification of cells and organelles using computer vision on cytology slides. We analyzed papers published in the last 4 years. The initial search was executed in September 2020 and resulted in 431 articles. After applying the inclusion/exclusion criteria, 157 papers remained, which we analyzed to build a picture of the tendencies and problems present in this research area, highlighting the computer vision methods, staining techniques, evaluation metrics, and the availability of the used datasets and computer code. As a result, we identified that the most used methods in the analyzed works are deep learning-based (70 papers), while fewer works employ classic computer vision only (101 papers). The most recurrent metric used for classification and object detection was the accuracy (33 papers and 5 papers), while for segmentation it was the Dice Similarity Coefficient (38 papers). Regarding staining techniques, Papanicolaou was the most employed one (130 papers), followed by H&E (20 papers) and Feulgen (5 papers). Twelve of the datasets used in the papers are publicly available, with the DTU/Herlev dataset being the most used one. We conclude that there still is a lack of high-quality datasets for many types of stains and most of the works are not mature enough to be applied in a daily clinical diagnostic routine. We also identified a growing tendency towards adopting deep learning-based approaches as the methods of choice.

Emissão de laudos eletrocardiográficos a distância: experiência da rede catarinense de telemedicina / Issuing electrocardiographic reports remotely: experience of the telemedicine network of Santa Catarina

FUNDAMENTO: Cresce o uso da Telemedicina, especialmente no envio e na avaliação de eletrocardiograma (ECG). É um procedimento de baixo custo, com alto potencial de salvar vidas. OBJETIVO: Descrever a forma de análise sistemática e o perfil do usuário da Rede Catarinense de Telemedicina quando o laudo era emitido de forma livre. MÉTODOS: Estudo observacional, transversal, determinando as associações entre características dos pacientes e diagnósticos eletrocardiográficos emitidos, dentre os usuários da Rede Catarinense de Telemedicina quando o laudo era fornecido de forma livre. Esse sistema estava conectado a 287 cidades de Santa Catarina, os exames eram feitos nos locais de origem e emitidos os laudos em três hospitais terciários. Entre 2005 e 2010, os laudos eram emitidos de forma livre e foi criado um método probabilístico para análise dos dados. Um cardiologista experiente avaliou todos os ECG para aferir a chance de anormalidade. RESULTADOS: Foram avaliados 243.363 ECG. A maioria (58%) foi realizada em pacientes com mais de 50 anos e proveniente da atenção primária (72%). Houve diferenças de frequência por região, parcialmente relacionado com número de cardiologistas/região (r = -0,551), com a distância dos centros terciários e com possíveis diferenças de aceitação do método. Cerca de 53% do ECG foram anormais, com maior frequência quanto maior a idade (r2 = 0,8166), e com diferenças regionais também significantes (p < 0,0001). CONCLUSÃO: Foi construído um sistema de análise dos dados integrando termos prevalentes, análise probabilística e dicionários especializados. O sistema tem atendido uma parcela significativa da população catarinense, principalmente idosos, da rede básica e de regiões remotas do estado.

BACKGROUND: A growing use of telemedicine has been observed, especially as regards the sending and evaluation of electrocardiograms (ECG); this is a low-cost procedure with a high potential to save lives. OBJECTIVES: To describe the form of systematic analysis and user profile of the Telemedicine Network of Santa Catarina during the time when the report was issued freely. METHODS: Observational cross-sectional study determining the associations between patient characteristics and electrocardiographic diagnoses issued among users of the Telemedicine Network of Santa Catarina during the time when the report was issued freely. This system was connected to 287 cities in Santa Catarina; the tests were done in the places of origin and the reports were issued in three tertiary-care hospitals. From 2005 to 2010 the reports were issued freely and a probabilistic method for data analysis was created. An experienced cardiologist evaluated all ECGs to assess the chances of abnormality. RESULTS: 243,363 ECGs were evaluated. The majority (58%) was performed on patients older than 50 years from primary care services (72%). There were differences in the frequency per region; this was partly related to the number of cardiologists/region (r = -0.551), to the distance from tertiary-care centers and potential differences of acceptance of the method. Approximately 53% of the ECGs were abnormal with greater frequency with increasing age (r² = 0.8166) and with significant regional differences (p < 0.0001). CONCLUSIONS: We built a data analysis system integrating prevalent terms, probabilistic analysis and specialized dictionaries. The system has covered a significant portion of the population of Santa Catarina, mainly elderly patients from the network of primary healthcare centers and remote regions of the State.

Reconhecimento e quantificação de expressão de imunoistoquímica empregando aprendizado de métricas de distância / Recognition and quantification of immunohistochemistry expression distance metric learning

Este artigo apresenta uma abordagem de segmentação para o reconhecimento e quantificação de expressão de imunoistoquímica (IHC) através do aprendizado de uma métrica de distância. Este método é baseado em duas etapas: treinamento e segmentação. A etapa de treinamento é realizada pela seleção supervisionada de algumas áreas típicas de expressão de IHC na imagem. Nesta etapa o padrão esperado de IHC é estatisticamente caracterizado, onde ocorre o aprendizado da métrica de distância e um espaço característico é modelado. Através desse espaço são obtidos os mapas de similaridade para cada imagem de IHC, com os níveis de intensidade correspondendo ao grau da reação do biomarcador sobre o tecido. A etapa de segmentação é guiada por um parâmetro de escala que controla a quantidade de áreas marcadas com base nos valores de intensidade dos mapas de similaridade. O método é baseado no aprendizado da distância de Mahalanobis para produzir um espaço característico, para posteriormente ser utilizado na distinção entre marcações positivas de expressão de IHC e tecidos normais, bem como quantificar o grau de intensidade da reação. Os resultados obtidos pelo método proposto foram comparados com a classificação linear no espaço de cores HSV (Hue, Saturation and Value) utilizando diferentes categorias de biomarcadores. Os resultados mostram que os limites da fronteira da distribuição dos padrões são mais bem definidos no método proposto, permitindo assim uma melhor discriminação entre tecidos normais e expressão de IHC.

This paper presents a segmentation approach to the recognition and quantification of immunohistochemistry (IHC) expression employing a distance metric learning method. This method is based in a two-step procedure, training and segmentation. The training step is performed by the supervised selection of a few IHC typical stained areas on image. In that step the desired IHC pattern is statistically characterized, where a distance metric is learned and a featured space is created. With this space, similarity maps are obtained by each IHC image with its intensity levels corresponding to degrees of reaction provided by the biomarker over the tissue. The segmentation step is guided by a scale-space parameter that controls the amount of labeled areas based on intensity values of the similarity maps. This method learns a Mahalanobis distance metric to produce a featured space used to distinguish between IHC positive staining and normal tissues, as well as quantifying the reaction intensity degrees. The results obtained by the proposed method were compared to the linear classification on HSV (Hue, Saturation and Value) color space using different biomarkers categories. The comparison results show that the boundary limits of the pattern distributions are better defined in the proposed method, allowing better discrimination between normal tissues and IHC expression.

White matter fiber tracking computation based on diffusion tensor imaging for clinical applications.

Fiber tracking allows the in vivo reconstruction of human brain white matter fiber trajectories based on magnetic resonance diffusion tensor imaging (MR-DTI), but its application in the clinical routine is still in its infancy. In this study, we present a new software for fiber tracking, developed on top of a general-purpose DICOM (digital imaging and communications in medicine) framework, which can be easily integrated into existing picture archiving and communication system (PACS) of radiological institutions. Images combining anatomical information and the localization of different fiber tract trajectories can be encoded and exported in DICOM and Analyze formats, which are valuable resources in the clinical applications of this method. Fiber tracking was implemented based on existing line propagation algorithms, but it includes a heuristic for fiber crossings in the case of disk-shaped diffusion tensors. We successfully performed fiber tracking on MR-DTI data sets from 26 patients with different types of brain lesions affecting the corticospinal tracts. In all cases, the trajectories of the central spinal tract (pyramidal tract) were reconstructed and could be applied at the planning phase of the surgery as well as in intraoperative neuronavigation.