Non-local attention and multi-task learning based lung segmentation in chest X-ray / 生物医学工程学杂志

Precise segmentation of lung field is a crucial step in chest radiographic computer-aided diagnosis system. With the development of deep learning, fully convolutional network based models for lung field segmentation have achieved great effect but are poor at accurate identification of the boundary and preserving lung field consistency. To solve this problem, this paper proposed a lung segmentation algorithm based on non-local attention and multi-task learning. Firstly, an encoder-decoder convolutional network based on residual connection was used to extract multi-scale context and predict the boundary of lung. Secondly, a non-local attention mechanism to capture the long-range dependencies between pixels in the boundary regions and global context was proposed to enrich feature of inconsistent region. Thirdly, a multi-task learning to predict lung field based on the enriched feature was conducted. Finally, experiments to evaluate this algorithm were performed on JSRT and Montgomery dataset. The maximum improvement of Dice coefficient and accuracy were 1.99% and 2.27%, respectively, comparing with other representative algorithms. Results show that by enhancing the attention of boundary, this algorithm can improve the accuracy and reduce false segmentation.

Robust pulmonary segmentation for chest radiography, combining enhancement, adaptive morphology and innovative active contours

Abstract Introduction Statistical data reveal that approximately 140 million radiological exams are performed annually in Brazil. These exams are designed to detect and to analyze fractures, caused by different types of trauma; as well as, to diagnose pathologies such as pulmonary diseases. For better visualization of those lesions or abnormalities, methods of image segmentation can be implemented. Such methods lead to the separation of the region of interest, which allows extracting the characteristics and anomalies of the desired tissue. However, the methods developed by researchers in this area still have restrictions. Consequently, we present an automatic pulmonary segmentation approach that overcomes these constraints. Methods This method is composed of a combination of Discrete Wavelet Packet Frame (DWPF), morphological operations and Gradient Vector Flow (GVF). The methodology is divided into four steps: Pre-processing - the original image is enhanced by discrete wavelet; Processing - where occurs a combination of the Otsu threshold with a series of morphological operations in order to identify the pulmonary object; Post-processing - an innovative form of using GVF improves the binary information of pulmonary tissue, and; Evaluation - the segmented images were evaluated for accuracy of detection the pulmonary region and border. Results The evaluation was carried out by segmenting 247 digital X-ray challenging images of the thorax human. The results show high for values of Overlap (97,63% ± 3.34%), and Average Contour Distance (0.69mm ± 0.95mm). Conclusion The results allow verifying that the proposed technique is robust and more accurate than other methods of lung segmentation, besides being a fully automatic method of lung segmentation.

Modelo de Contorno Ativo Crisp Adaptativo 2D aplicado na segmentação dos pulmões em imagens de TC do tórax de voluntários sadios e pacientes com enfisema pulmonar / Adaptive 2D Crisp Active Contour Model applied to lung segmentation in CT images of the thorax of healthy volunteers and patients with pulmonary emphysema

INTRODUÇÃO: Dentre as doenças que afetam a população mundial, destaca-se a preocupação com a Doença Pulmonar Obstrutiva Crônica (DPOC), que, segundo a Organização Mundial de Saúde, pode se constituir na terceira causa de morte mais importante em todo mundo no ano de 2030. Visando contribuir com o auxílio ao diagnóstico médico, esta pesquisa centraliza seus esforços na etapa de segmentação dos pulmões, visto que esta é a etapa básica de sistema de Visão Computacional na area de pneumologia. MÉTODOS: Este trabalho propõe um novo método de segmentação dos pulmões em imagens de Tomografia Computadorizada (TC) do tórax chamado de Método de Contorno Ativo (MCA) Crisp Adaptativo 2D. Este MCA consiste em traçar automaticamente uma curva inicial dentro dos pulmões, que se deforma por iterações sucessivas, minimizando energias que atuam sobre a mesma, deslocando-a até as bordas do objeto. O MCA proposto é o resultado do aperfeiçoamento do MCA Crisp desenvolvido previamente, visando aumentar a sua exatidão, diminuindo o tempo de análise e reduzindo a subjetividade na segmentação e análise dos pulmões dessas imagens pelos médicos especialistas. Este método por iterações sucessivas de minimização de sua energia, segmenta de forma automática os pulmões em imagens de TC do tórax. RESULTADOS: Para sua validação, o MCA Crisp Adaptativo é comparado com os MCAs THRMulti, THRMod, GVF, VFC, Crisp e também com o sistema SISDEP, sendo esta avaliação realizada utilizando como referência 24 imagens, sendo 12 de pacientes com DPOC e 12 de voluntários sadios, segmentadas manualmente por um pneumologista. Os resultados obtidos demonstram que o método proposto é superior aos demais. CONCLUSÃO: Diante dos resultados obtidos, pode-se concluir que este método pode integrar sistemas de auxílio ao diagnóstico médico na área de Pneumologia.

INTRODUCTION: Among the diseases that affect the world's population, there is concern about Chronic Obstructive Pulmonary Disease (COPD), that, according to the World Health Organization, could be the leading cause of death worldwide by the year 2030. Aiming to contribute to aid medical diagnosis, this research focuses its efforts on the segmentation of the lungs, since this is the basic step system in the area of Computer Vision pulmonology. METHODS: This paper proposes a new method for segmentation of lung images in Computed Tomography (CT) of the chest called Active Contour Method (MCA) Crisp Adaptive 2D. This MCA is to draw a curve starting inside an object of interest. This curve is deformed by successive iterations, minimizing energies that act on it, moving it to the edges of the object. The MCA is the improvement of the proposed MCA Crisp previously developed, aiming to increase the accuracy, decreasing analysis time and reducing the subjectivity in the segmentation and analysis of the lungs of these images by pulmonologists. This method is automatically initialized in the lungs and on successive iterations to minimize this energy, this MCA automatically targets the lungs in chest CT images. RESULTS: To evaluate the proposed method, the MCA Adaptive Crisp is compared with MCAs THRMulti, THRMod, GVF, VFC, Crisp and also with the system SISDEP, this assessment is performed using reference images 24, 12 COPD patients and 12 volunteers healthy, manually segmented by a pulmonologist. The results show that the proposed method is superior to the others. CONCLUSION: Based on the results, it can be concluded that this method can integrate systems aid in the medical diagnosis of Pulmonology.

Modelo de contorno ativo crisp: nova técnica de segmentação dos pulmões em imagens de TC / Active contour modes crisp: new technique for segmentation the lungs in CT images

Este trabalho propõe um novo método de contornos ativos (MCA), chamado de MCA Crisp, e o avalia na segmentação dos pulmões em imagens de Tomografia Computadorizada (TC). O MCA consiste em traçar uma curva inicial em torno ou dentro de um objeto de interesse. Esta curva se deforma, conforme algumas energias que atuam sobre a mesma, deslocando-a até as bordas do objeto. Este processo é realizado por iterações sucessivas de minimização de uma dada energia, associada à curva. Aplicando MCAs descritos na literatura na segmentação dos pulmões em imagens de TC, constatam-se limitações. Neste contexto, propõe‑-se o MCA Crisp para suprir tais limitações, propondo uma inicialização automática e uma nova energia externa baseada em regras e nas densidades radiológicas pulmonares. Realiza-se uma comparação entre os MCAs Tradicional, Balão, GVF e o método proposto para demonstrar a superioridade do novo método. Em seguida, para validar o MCA Crisp realiza-se uma análise qualitativa junto a um médico especialista na área de Pneumologia do Hospital Universitário Walter Cantídio da Universidade Federal do Ceará. Nesta análise, são utilizados 100  pulmões em imagens de TC. A eficiência da segmentação foi avaliada em  5 categorias, obtendo os seguintes resultados:   73% ótimas, sem falhas, 20% aceitáveis, com pequenos erros, 7% razoáveis, com erros grosseiros e  0% ruim, segmentando apenas uma pequena parte do pulmão, e  0% péssima, obtendo uma segmentação totalmente errada. Conclui-se que o MCA Crisp é um método útil para segmentar os pulmões em imagens de TC e com potencial para integrar sistemas que auxiliem o diagnóstico médico.

This paper proposes a new Active Contour Model (ACM), called ACM Crisp, and evaluates the segmentation of lungs in computed tomography (CT) images. An ACM draws a curve around or within the object of interest. This curve changes its shape, when some energy acts on it and moves towards the edges of the object. This process is performed by successive iterations of minimization of a given energy, associated with the curve. The ACMs described in the literature have limitations when used for segmentations of CT lung images. The ACM Crisp model overcomes these limitations, since it proposes automatic initiation and new external energy based on rules and radiological pulmonary densities. The paper compares other ACMs with the proposed method, which is shown to be superior. In order to validate the algorithm a medical expert in the field of Pulmonology of the Walter Cantídio University Hospital from the Federal University of Ceará carried out a qualitative analysis. In these analyses  100 CT lung images were used. The segmentation efficiency was evaluated into  5 categories with the following results for the ACM Crisp: 73% excellent, without errors, 20% acceptable, with small errors, and  7% reasonable, with large errors, 0% poor, covering only a small part of the lung, and  0% very bad, making a totally incorrect segmentation. In conclusion the ACM Crisp is considered a useful algorithm to segment CT lung images, and with potential to integrate medical diagnosis systems.