RESUMO
Although it has high resolution for soft tissues,magnetic resonance imaging(MRI)is not the standard for chest imaging,which results in an insufficient amount of expert-annotated MRI data.Therefore,CT image is usually converted into MRI image.To overcome the difficulty of obtaining the corresponding modal CT and MRI images,a CSCGAN model with CycleGAN as the framework is proposed based on the structural characteristics of generative adversarial networks.Considering the possibility of mode collapse in CycleGAN,StyleGan2 which can control the style and feature details of the synthetic image and realize the synthesis of high-resolution images is integrated into CycleGAN for reconstructing the generator.A noise module is introduced to reduce external interference.In addition,in order to prevent the loss of tumors during conversion,the discriminator structure of the network is modified,and a mixed attention mechanism is added.Experimental results show that compared with the images generated by other methods,those generated by the proposed model are improved in Dice similarity coefficient,Hausdorff distance,volume ratio and mean intersection over union,indicating that the proposed method can effectively realize the mode conversion of liver tumor images,and that the generated data can improve the segmentation accuracy.
RESUMO
A U-Net incorporating improved Transformer and convolutional channel attention module is designed for biventricular segmentation in MRI image.By replacing the high-level convolution of U-Net with the improved Transformer,the global feature information can be effectively extracted to cope with the challenge of poor segmentation performance due to the complex morphological variation of the right ventricle.The improved Transformer incorporates a fixed window attention for position localization in the self-attention module,and aggregates the output feature map for reducing the feature map size;and the network learning capability is improved by increasing network depth through the adjustment of multilayer perceptron.To solve the problem of unsatisfactory segmentation performance caused by blurred tissue edges,a feature aggregation module is used for the fusion of multi-level underlying features,and a convolutional channel attention module is adopted to rescale the underlying features to achieve adaptive learning of feature weights.In addition,a plug-and-play feature enhancement module is integrated to improve the segmentation performance which is affected by feature loss due to channel decay in the codec structure,which guarantees the spatial information while increasing the proportion of useful channel information.The test on the ACDC dataset shows that the proposed method has higher biventricular segmentation accuracy,especially for the right ventricle segmentation.Compared with other methods,the proposed method improves the DSC coefficient by at least 2.83%,proving its effectiveness in biventricular segmentation.
RESUMO
In response to the current situation and teaching status of the medical image processing course on the background of"new medical science",a teaching software which is highly compatible with the teaching process of medical image processing is developed.The teaching software allows for linear grayscale transformation,windowing display,scaling,rotation,mirroring,median filtering,differential sharpening,edge detection,histogram acquisition,and histogram equalization of medical images.Additionally,it enables parameter adjustments within a certain range for linear grayscale transformation,windowing display,scaling,rotation,median filtering,differential sharpening,and edge detection.Meanwhile,it employs different algorithms to achieve the scaling of medical images.The teaching software is used in the theoretical and experimental teaching of medical image processing courses at Baotou Medical College.It can improve students'initiative and enthusiasm in learning,strengthen their understanding of the examination points for radiology technicians,lay a solid foundation for subsequent courses,and ultimately achieve the goal of in-depth integration of"Medical Engineering"and"Medical Science"in the medical image technology major at Baotou Medical College under the background of"new medical science".
RESUMO
Purpose To evaluate the feasibility of cardiac magnetic resonance fractal analysis in evaluating left ventricular trabecular complexity in hypertrophic cardiomyopathy(HCM),and to study the degree of left ventricular trabecular complexity in HCM and the relationship between excessive trabecular complexity and cardiac function.Materials and Methods From August 2020 to December 2022,a total of 80 patients with HCM from the Second Affiliated Hospital of Nanchang University were retrospectively analyzed.Additionally,80 healthy volunteers were recruited as the control group.Left ventricular functional parameters and fractal dimension(FD)of left ventricular trabecular myocardium were measured.The differences of mean global FD,max basal FD and max apical FD were compared between the HCM group and the control group,the correlation between FDs and cardiac function parameters was evaluated.The diagnostic efficiency of mean global FD,max apical FD and max basal FD was analyzed via receiver operating characteristic curve.Results The mean global FD of HCM group was significantly higher than that of normal group,and the difference was statistically significant(1.303±0.047 vs.1.229±0.026;t=-12.387,P<0.001).Mean global FD showed the best performance in differentiating HCM from normal control group.The optimal cut-off value for the diagnosis of HCM was 1.251,with the area under curve of 0.933(95%CI 0.896-0.969).Mean global FD was positively correlated with maximum wall thickness and left ventricular mass index(r=0.686,0.687,P<0.001),and max apical FD was positively correlated with left ventricular ejection fraction(r=0.520,P<0.001).Conclusion The FD obtained by cardiac magnetic resonance fractal analysis technique is reproducible and has definite value in the diagnosis of HCM,with association with the structure and function of left heart.
RESUMO
Objective:To investigate whether the image quality of total-body PET/CT (TB PET/CT) with 1 min acquisition can meet the clinical diagnostic requirements.Methods:From May 2019 to September 2021, a total of 90 malignant tumor patients (60 males, 30 females, age 31-86 years) with primary lesions confirmed by pathological diagnosis in Zhongshan Hospital, Fudan University were respectively analyzed. All patients underwent conventional PET/CT (C PET/CT) scan with conventional clinical acquisition and TB PET/CT scan with 1 min acquisition after injecting 18F-FDG in random order. Paired t test or Wilcoxon signed rank test was used to analyze the image quality of these two scans. Results:SUV max of primary lesions in TB PET/CT group was significantly higher than that in C PET/CT group (15.9(7.9, 24.6) vs 12.5(5.8, 16.6); z=8.14, P<0.001), so were signal-to-noise ratio (SNR) of the blood pool, liver, muscles (9.3±3.0, 11.4(9.5, 14.2), 8.3(7.3, 10.1) vs 6.2±1.7, 9.4(7.7, 11.8), 6.0(4.9, 7.1)), tumor-to-blood pool ratio (TBR) (9.3(4.3, 14.8) vs 8.5(4.3, 11.1)), tumor-to-liver ratio (TLR) (6.7(3.0, 10.4) vs 6.1(2.9, 7.7)), tumor-to-muscle ratio (TMR) (23.2(11.5, 38.0) vs 18.3(9.6, 26.6); t=9.36, z values: 4.44-7.40, all P<0.001). Conclusion:The image quality of TB PET/CT scan with 1 min acquisition can meet the diagnostic requirements, and is better than the C PET/CT image quality with conventional clinical acquisition.
RESUMO
Objective:To compare the imaging quality and metabolic quantitative parameters of pulmonary nodules between Q. Flex whole information five-dimensional (5D) and conventional three-dimensional (3D) PET/CT imaging for clinical evaluation.Methods:Fifty-four patients (30 males, 24 females, age: 60(42, 75) years; 78 solid pulmonary nodules (maximum diameter≤3 cm) with abnormal uptake of 18F-FDG) from Tianjin Cancer Hospital Airport Hospital between June 2022 and August 2022 were enrolled in this retrospective study. All patients underwent 5D scanning and 3D, 5D reconstruction. Image quality scores, signal-to-noise ratio (SNR), SUV max, SUV mean and metabolic tumor volume (MTV) of pulmonary nodules of 5D group and 3D group were evaluated and compared with χ2 test and Wilcoxon signed rank test. Correlation of quantitative parameters between 2 groups were analyzed by using Spearman rank correlation analysis. Results:Thirty-five of 78(45%) pulmonary nodules with image quality score≥4 were found in 5D group, which were more than those in 3D group (22/78(28%); χ2=4.67, P=0.031). Meanwhile, SNR, SUV max, SUV mean, and MTV were significantly positively correlated between the 2 groups ( rs values: 0.86, 0.86, 0.85, and 0.95, all P<0.001). SNR, SUV max and SUV mean of pulmonary nodules in 5D group were significantly higher than those in 3D group, which were 37.46(18.42, 62.00) vs 32.72(16.97, 54.76) ( z=-4.07, P<0.001), 9.71(5.48, 13.82) vs 8.96(4.82, 12.63) ( z=-3.05, P<0.001) and 6.30(3.39, 8.94) vs 5.61(2.99, 7.63)( z=-4.07, P<0.001) respectively. MTV of pulmonary nodules in 5D group was significantly lower than that in 3D group, which was 1.72(0.66, 2.74) cm 3vs 1.98(1.06, 4.63) cm 3 ( z=-7.13, P<0.001). Quantitative parameters of lower lung field and nodules with maximum diameters of >10 mm and ≤20 mm based on 5D scanning changed most significantly compared with those based on 3D scanning ( z values: from -5.23 to -2.48, all P<0.05). Conclusion:Q. Flex 5D PET significantly improves the quantitative accuracy of SUV and MTV of pulmonary nodules, and the improvement of image quality is substantial without increasing the radiation dose, which has clinical practical value.
RESUMO
Dual-tracer PET imaging can provide more comprehensive clinical information for disease diagnosis. How to achieve dual-tracer imaging through single imaging session is one of the hot topics in the field of nuclear medicine. The key issues in achieving dual-tracer imaging with a single scan is the accurate separation of single-tracer information from the mixed tracer signal. In the last two decades, separation methods for different tracers have been proposed, and the injection interval between two tracers was reduced. This paper summarizes recent dual-tracer separation methods. The basic principles, features and advantages of separation methods are analyzed. Besides, the outlook on the future development of dual-tracer separation is also discussed in the present review.
RESUMO
The incidence of malnutrition in surgical patients is high and affects the clinical outcome of patients. Through the extraction and analysis of image data, radiomics can assess changes in the composition of the body, such as skeletal muscle and fat, and it demonstrates tremendous potential in nutritional screening, assessment, diagnosis, and evaluation of treatment effects, emerging as a crucial evaluation tool in nutritional therapy for surgical patients. Furthermore, radiomics can predict patients' clinical outcomes, providing more precise treatment plans. Therefore, the application of radiomics should be fully emphasized in nutritional therapy for surgical patients to promote their enhanced recovery. With the continuous development and improvement of radiomics technology in the future, its application in nutritional therapy for surgical patients is expected to become more extensive and profound, bringing better treatment outcomes and quality of life to patients.
RESUMO
Objective:To explore the application value of optical surface monitoring system (OSMS) volume rendering technique (VRT) body surface imaging in intensity-modulated radiotherapy for thoracic tumors.Methods:A retrospective case series study was performed. The clinical data of 65 patients with thoracic tumors treated with intensity-modulated radiotherapy at Tongji Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology from September 2021 to October 2022 were retrospectively analyzed. In the first treatment,after cone-beam computed tomography (CBCT) scan and correction, VRT body surface images were obtained by using OSMS. In subsequent treatment, the VRT image was used as the benchmark and the 6-dimensional bed was automatically positioned to record the 6-dimensional bed positioning value. The CBCT scan was performed to record the translation and rotation errors of left-right direction (X-axis), head-foot direction (Y-axis) and front-rear direction (Z-axis). After the calibration of the 6-dimensional automatic bed shifting, the new real-time deltas (RTD) value of digital imaging and communications in medicine (DICOM) body surface image was recorded, and the new VRT image was obtained. CBCT registration error value was defined as VRT image-guided setup error. The sum of CBCT registration error value and moving bed movement value was defined as the body surface marker line-guided setup error. The sum of CBCT registration error value and the recorded DICOM image RTD value was defined as the theoretical error of DICOM image-guided setup. The advantages and disadvantages of VRT image, body surface marker line and DICOM image-guided setup were compared and analyzed.Results:There were 42 males and 23 females in 65 patients with thoracic tumors, and the age [ M ( Q1, Q3)] was 58 years (51 years, 64 years). The linear errors [ M ( Q1, Q3)] of VRT image-guided setup in X, Y and Z axes were 0.6 mm (0.3 mm, 1.2 mm), 1.2 mm (0.5 mm, 2.4 mm) and 1.1 mm (0.5 mm, 1.9 mm); and the rotational errors were 0.4° (0.1°, 0.7°), 0.4° (0.1°, 0.6°) and 0.4° (0.2°, 0.6°). The linear errors of the marker line-guided setup were 1.6 mm (0.9 mm, 2.6 mm), 2.2 mm (1.1 mm, 3.8 mm) and 1.0 mm (0.4 mm, 1.8 mm); and the rotational errors were 0.7° (0.3°, 1.2°), 0.5° (0.2°, 0.8°) and 0.5° (0.2°, 0.8°). The linear errors of the DICOM image-guided positioning were 1.1 mm (0.6 mm, 1.9 mm), 2.1 mm (1.0 mm, 3.4 mm) and 1.3 mm (0.6 mm, 3.1 mm), and the rotational errors were 0.6° (0.2°, 1.1°), 0.7° (0.3°, 1.1°), 0.7° (0.2°, 1.1°). Compared with the marker line-guided setup, except for Z-axis linear error ( P = 0.218), the VRT-guided setup errors were low (all P < 0.001). Compared with the DICOM imaging-guided setup, the VRT image-guided setup linear error and rotational error in X-, Y- and Z-axis were low (all P < 0.01). Conclusions:VRT image-guided setup is superior to traditional body surface marker setup and DICOM imaging setup; OSMS VRT body surface imaging can effectively improve the setup accuracy and stability of intensity-modulated radiotherapy for thoracic tumors, and reduce the setup errors.
RESUMO
Objective To observe the value of multi-slice spiral CT(MSCT)post-processing technologies for diagnosing otosclerosis.Methods Clinical data and original axial plain MSCT of 47 patients with otosclerosis(92 ears)and 65 patients with non-otosclerosis hearing impairment(79 ears)were retrospectively enrolled.MSCT post-processing images,including multi-planar reformation(MPR)of stapes and cochleas and curved planar reformation(CPR)of ossicular chains were obtained.The diagnostic value of original MSCT images alone and raw data of MSCT combing with post-processing images for diagnosing otosclerosis were compared.Results Otosclerosis was correctly diagnosed in 66 ears according to original MSCT images alone,but in 89 ears combined with MSCT post-processing images.The sensitivity of original MSCT images alone and combined with MSCT post-processing images was 71.74％and 96.74％,respectively,and the diagnostic accuracy was 81.29％and 96.49％,respectively,those of the latter were both higher than of the former(both P＜0.05),which had specificities being not significantly different(92.41％vs.96.20％,P＞0.05).Conclusion Combining with post-processing technologies could increase the sensitivity and accuracy of MSCT for diagnosing otosclerosis.
RESUMO
Objective To observe the value of SimGrid(SG)and S-Enhance(SE)for improving image quality of low-dose X-ray films in children.Methods Data of 344 children in intensive care unit who underwent 410 times bedside X-ray examinations,including 290 times of chest X-ray,51 of abdominal X-ray and 69 of chest and abdominal combined X-ray were enrolled.SG and SE were respectively used for post-processing,and the quality of post-processed images were analyzed.Results Among 410 SG post-processing images,250 images were classified as 2-point,147 as 1-point and 13 as 0-point.SG could significantly improve image quality of children≥1 year and body mass≥10 kg(all P＜0.05),with better ability for displaying bones,trachea,peripheral blood vessels,foreign objects,psoas major muscle and intestinal gas(all P＜0.05).Among 410 SE post-processing images,250 images were classified as 2-point,58 as 1-point and 102 of 0-point.SE could significantly improve image quality of children≥0.5 years and with body mass＞4 kg(all P＜0.05),with better ability for displaying bones,trachea,large blood vessels,peripheral vessels,heart posterior blood vessels and foreign objects(all P＜0.05).Conclusion SG could significantly improve displaying of bones,trachea,peripheral blood vessels,foreign objects,psoas major muscle and intestinal gas in children≥1 year and body mass≥10 kg,while SE could improve displaying of bones,trachea,large blood vessels,peripheral blood vessels,heart posterior blood vessels and foreign objects in children aged≥0.5 years and body mass＞4 kg on low-dose X-ray films.
RESUMO
Objective@#To research the effectiveness of deep learning techniques in intelligently diagnosing dental caries and periapical periodontitis and to explore the preliminary application value of deep learning in the diagnosis of oral diseases@*Methods@#A dataset containing 2 298 periapical films, including healthy teeth, dental caries, and periapical periodontitis, was used for the study. The dataset was randomly divided into 1 573 training images, 233 validation images, and 492 test images. By comparing various neural network models, the MobileNetV3 network model with better performance was selected for dental disease diagnosis, and the model was optimized by tuning the network hyperparameters. The accuracy, precision, recall, and F1 score were used to evaluate the model's ability to recognize dental caries and periapical periodontitis. Class activation map was used to visualization analyze the performance of the network model@*Results@#The algorithm achieved a relatively ideal intelligent diagnostic effect with precision, recall, and accuracy of 99.42%, 99.73%, and 99.60%, respectively, and the F1 score was 99.57% for classifying healthy teeth, dental caries, and periapical periodontitis. The visualization of the class activation maps also showed that the network model can accurately extract features of dental diseases.@*Conclusion@#The tooth lesion detection algorithm based on the MobileNetV3 network model can eliminate interference from image quality and human factors and has high diagnostic accuracy, which can meet the needs of dental medicine teaching and clinical applications.
RESUMO
ABSTRACT Purpose: Evaluation of lid contour and marginal peak point changes to compare outcomes of external levator advancement and Müller's muscle conjunctival resection surgery in unilateral ptosis. Methods: We reviewed the charts of unilateral ptosis patients who underwent external levator advancement or Müller's muscle conjunctival resection. Eyelid contour analysis was conducted on preoperative and 6-month postoperative digital images. This was performed with the multiple margin reflex distances technique, measuring the vertical distance from a line intersecting the center of the pupil to the eyelid margin at 10 positions at 2 mm intervals. The marginal peak point changes were analyzed digitally using the coordinates of the peak point according to the pupil center. Each position's mean distance was compared preoperatively, postoperatively, and with the fellow eyelid. Results: Sixteen patients underwent external levator advancement and 16 patients had Müller's muscle conjunctival resection. The mean margin reflex distance was improved by both techniques (1.46 vs. 2.43 mm and 1.12 vs. 2.25 mm, p=0.008 and p=0.0001 respectively) and approached that of the fellow eyelid (2.43 vs. 2.88 and 2.25 vs. 2.58 mm, p=0.23 and p=0.19, respectively). However, statistically significant lid margin elevation was limited to between the N6 and T6 points in the external levator advancement group. Whereas, significant elevation was achieved along the whole lid margin in the Müller's muscle conjunctival resection group. The marginal peak point was shifted slightly laterally in the external levator advancement group (p=0.11). Conclusions: Both techniques provide effective lid elevation, however, the external levator advancement's effect lessens toward the canthi while Müller's muscle conjunctival resection provides more uniform elevation across the lid margin. The margin reflex distance alone is not sufficient to reflect contour changes.
RESUMO
ABSTRACT Background: Colorectal cancer is one of the most prevalent pathologies worldwide whose prognosis is linked to early detection. Colonoscopy is the gold standard for screening, and diagnosis is usually made histologically from biopsies. Aiming to reduce the inspection and diagnostic time as well as the biopsies and resources involved, other techniques are being promoted to conduct accurate in vivo colonoscopy assessments. Optical biopsy aims to detect normal and neoplastic tissues analysing the autofluorescence spectrum based on the changes in the distribution and concentration of autofluorescent molecules caused by colorectal cancer. Therefore, the autofluorescence contribution analysed by image processing techniques could be an approach to a faster characterization of the target tissue. Objective: Quantify intensity parameters through digital processing of two data sets of three-dimensional widefield autofluorescence microscopy images, acquired by fresh colon tissue samples from a colorectal cancer murine model. Additionally, analyse the autofluorescence data to provide a characterization over a volume of approximately 50 µm of the colon mucosa for each image, at second (2nd), fourth (4th) and eighth (8th) weeks after colorectal cancer induction. Methods: Development of a colorectal cancer murine model using azoxymethane/dextran sodium sulphate induction, and data sets acquisition of Z-stack images by widefield autofluorescence microscopy, from control and colorectal cancer induced animals. Pre-processing steps of intensity value adjustments followed by quantification and characterization procedures using image processing workflow automation by Fiji's macros, and statistical data analysis. Results: The effectiveness of the colorectal cancer induction model was corroborated by a histological assessment to correlate and validate the link between histological and autofluorescence changes. The image digital processing methodology proposed was then performed on the three-dimensional images from control mice and from the 2nd, 4th, and 8th weeks after colorectal cancer chemical induction, for each data set. Statistical analyses found significant differences in the mean, standard deviation, and minimum parameters between control samples and those of the 2nd week after induction with respect to the 4th week of the first experimental study. This suggests that the characteristics of colorectal cancer can be detected after the 2nd week post-induction. Conclusion: The use of autofluorescence still exhibits levels of variability that prevent greater systematization of the data obtained during the progression of colorectal cancer. However, these preliminary outcomes could be considered an approach to the three-dimensional characterization of the autofluorescence of colorectal tissue, describing the autofluorescence features of samples coming from dysplasia to colorectal cancer.
RESUMO Contexto: O câncer colorretal é uma das patologias mais prevalentes em todo o mundo, cujo prognóstico está ligado à detecção precoce. A colonoscopia é o padrão ouro para triagem, e o diagnóstico geralmente é feito histologicamente a partir de biópsias. Visando reduzir o tempo de inspeção e diagnóstico, bem como as biópsias e recursos envolvidos, outras técnicas estão sendo promovidas para realizar avaliações precisas de colonoscopia in vivo. A biópsia óptica visa detectar tecidos normais e neoplásicos analisando o espectro de autofluorescência com base nas mudanças na distribuição e concentração de moléculas autofluorescentes causadas pelo câncer colorretal. Portanto, a contribuição da autofluorescência analisada por técnicas de processamento de imagem poderia ser uma abordagem para uma caracterização mais rápida do tecido-alvo. Objetivo: Quantificar parâmetros de intensidade por meio do processamento digital de dois conjuntos de dados de imagens de microscopia de autofluorescência em campo amplo tridimensionais, adquiridas por amostras de tecido fresco de cólon de um modelo murino de câncer colorretal. Adicionalmente, analisar os dados de autofluorescência para fornecer uma caracterização em um volume de aproximadamente 50 µm da mucosa do cólon para cada imagem, na segunda (2ª), quarta (4ª) e oitava (8ª) semanas após a indução do câncer colorretal. Método: Desenvolvimento de um modelo murino de câncer colorretal usando indução de azoximetano/sulfato de sódio dextrano, e aquisição de conjuntos de dados de imagens Z-stack por microscopia de autofluorescência em campo amplo, de animais controle e induzidos ao câncer colorretal. Etapas de pré-processamento de ajustes de valores de intensidade seguidas por procedimentos de quantificação e caracterização usando automação de fluxo de trabalho de processamento de imagem por macros do Fiji, e análise estatística de dados. Resultados: A eficácia do modelo de indução de câncer colorretal foi corroborada por uma avaliação histológica para correlacionar e validar a ligação entre as mudanças histológicas e de autofluorescência. A metodologia de processamento digital de imagem proposta foi então realizada nas imagens tridimensionais de camundongos controle e das 2ª, 4ª e 8ª semanas após a indução química do câncer colorretal, para cada conjunto de dados. Análises estatísticas encontraram diferenças significativas nos parâmetros médios, desvio padrão e mínimos entre amostras de controle e aquelas da 2ª semana após a indução em relação à 4ª semana do primeiro estudo experimental. Isso sugere que as características do câncer colorretal podem ser detectadas após a 2ª semana pós-indução. Conclusão: O uso de autofluorescência ainda apresenta níveis de variabilidade que impedem uma maior sistematização dos dados obtidos durante a progressão do câncer colorretal. No entanto, esses resultados preliminares podem ser considerados uma abordagem para a caracterização tridimensional da autofluorescência do tecido colorretal, descrevendo as características de autofluorescência de amostras que vão da displasia ao câncer colorretal.
RESUMO
ABSTRACT Purpose: This study aimed to evaluate the classification performance of pretrained convolutional neural network models or architectures using fundus image dataset containing eight disease labels. Methods: A publicly available ocular disease intelligent recognition database has been used for the diagnosis of eight diseases. This ocular disease intelligent recognition database has a total of 10,000 fundus images from both eyes of 5,000 patients for the following eight diseases: healthy, diabetic retinopathy, glaucoma, cataract, age-related macular degeneration, hypertension, myopia, and others. Ocular disease classification performances were investigated by constructing three pretrained convolutional neural network architectures including VGG16, Inceptionv3, and ResNet50 models with adaptive moment optimizer. These models were implemented in Google Colab, which made the task straight-forward without spending hours installing the environment and supporting libraries. To evaluate the effectiveness of the models, the dataset was divided into 70%, 10%, and 20% for training, validation, and testing, respectively. For each classification, the training images were augmented to 10,000 fundus images. Results: ResNet50 achieved an accuracy of 97.1%; sensitivity, 78.5%; specificity, 98.5%; and precision, 79.7%, and had the best area under the curve and final score to classify cataract (area under the curve = 0.964, final score = 0.903). By contrast, VGG16 achieved an accuracy of 96.2%; sensitivity, 56.9%; specificity, 99.2%; precision, 84.1%; area under the curve, 0.949; and final score, 0.857. Conclusions: These results demonstrate the ability of the pretrained convolutional neural network architectures to identify ophthalmological diseases from fundus images. ResNet50 can be a good architecture to solve problems in disease detection and classification of glaucoma, cataract, hypertension, and myopia; Inceptionv3 for age-related macular degeneration, and other disease; and VGG16 for normal and diabetic retinopathy.
RESUMO Objetivo: Avaliar o desempenho de classificação de modelos ou arquiteturas de rede neural convolucional pré--treinadas usando um conjunto de dados de imagem de fundo de olho contendo oito rótulos de doenças diferentes. Métodos: Neste artigo, o conjunto de dados de reconhecimento inteligente de doenças oculares publicamente disponível foi usado para o diagnóstico de oito rótulos de doenças diferentes. O banco de dados de reconhecimento inteligente de doenças oculares tem um total de 10.000 imagens de fundo de olho de ambos os olhos de 5.000 pacientes para oito categorias que contêm rótulos saudáveis, retinopatia diabética, glaucoma, catarata, degeneração macular relacionada à idade, hipertensão, miopia, outros. Investigamos o desempenho da classificação de doenças oculares construindo três arquiteturas de rede neural convolucional pré-treinadas diferentes, incluindo os modelos VGG16, Inceptionv3 e ResNet50 com otimizador de Momento Adaptativo. Esses modelos foram implementados no Google Colab o que facilitou a tarefa sem gastar horas instalando o ambiente e suportando bibliotecas. Para avaliar a eficácia dos modelos, o conjunto de dados é dividido em 70% para treinamento, 10% para validação e os 20% restantes utilizados para teste. As imagens de treinamento foram expandidas para 10.000 imagens de fundo de olho para cada tal. Resultados: Observou-se que o modelo ResNet50 alcançou acurácia de 97,1%, sensibilidade de 78,5%, especificidade de 98,5% e precisão de 79,7% e teve a melhor área sob a curva e pontuação final para classificar a categoria da catarata (área sob a curva=0,964, final=0,903). Em contraste, o modelo VGG16 alcançou uma precisão de 96,2%, sensibilidade de 56,9%, especificidade de 99,2% e precisão de 84,1%, área sob a curva 0,949 e pontuação final de 0,857. Conclusão: Esses resultados demonstram a capacidade das arquiteturas de rede neural convolucional pré-treinadas em identificar doenças oftalmológicas a partir de imagens de fundo de olho. ResNet50 pode ser uma boa solução para resolver problemas na detecção e classificação de doenças como glaucoma, catarata, hipertensão e miopia; Inceptionv3 para degeneração macular relacionada à idade e outras doenças; e VGG16 para retinopatia normal e diabética.
RESUMO
Abstract Objective: To develop an automated co-registration system and test its performance, with and without a fiducial marker, on single-photon emission computed tomography (SPECT) images. Materials and Methods: Three SPECT/CT scans were acquired for each rotation of a Jaszczak phantom (to 0°, 5°, and 10° in relation to the bed axis), with and without a fiducial marker. Two rigid co-registration software packages-SPM12 and NMDose-coreg-were employed, and the percent root mean square error (%RMSE) was calculated in order to assess the quality of the co-registrations. Uniformity, contrast, and resolution were measured before and after co-registration. The NMDose-coreg software was employed to calculate the renal doses in 12 patients treated with 177Lu-DOTATATE, and we compared those with the values obtained with the Organ Level INternal Dose Assessment for EXponential Modeling (OLINDA/EXM) software. Results: The use of a fiducial marker had no significant effect on the quality of co-registration on SPECT images, as measured by %RMSE (p = 0.40). After co-registration, uniformity, contrast, and resolution did not differ between the images acquired with fiducial markers and those acquired without. Preliminary clinical application showed mean total processing times of 9 ± 3 min/patient for NMDose-coreg and 64 ± 10 min/patient for OLINDA/EXM, with a strong correlation between the two, despite the lower renal doses obtained with NMDose-coreg. Conclusion: The use of NMDose-coreg allows fast co-registration of SPECT images, with no loss of uniformity, contrast, or resolution. The use of a fiducial marker does not appear to increase the accuracy of co-registration on phantoms.
Resumo Objetivo: Desenvolver corregistro automático e testar seu desempenho com ou sem marcador fiducial em imagens de tomografia computadorizada de emissão de fóton único (SPECT). Materiais e Métodos: Três SPECT/CTs foram adquiridas para cada rotação de um simulador de Jaszczak em relação ao eixo da maca (0°, 5° e 10°), com e sem fiducial. Dois métodos de corregistro inelástico foram aplicados - SPM12 e NMDose-coreg -, e a porcentagem do erro quadrático médio (%RMSE) foi usada para analisar a qualidade do corregistro. Uniformidade, contraste e resolução foram medidos antes e após o corregistro. NMDose com corregistro automático foi usado para calcular a dose renal de 12 pacientes tratados com 177Lu-DOTATATE e comparado com OLINDA/EXM. Resultados: A marcação fiducial não modificou a qualidade do corregistro das imagens SPECT, medida pela %RMSE (p = 0,40). Não houve impacto na uniformidade, contraste e resolução após o corregistro de imagens adquiridas com ou sem fiduciais. Aplicação clínica preliminar mostrou tempo total de processamento de 9 ± 3 min/paciente para NMDose e 64 ± 10 min/paciente para OLINDA/EXM, com alta correlação entre ambos, apesar de menor dose renal em NMDose. Conclusão: NMDose-coreg permite o corregistro rápido de imagens SPECT, sem perda de uniformidade, contraste ou resolução. O uso da marcação fiducial não aumentou a precisão do corregistro em fantomas.
RESUMO
Fundamento: A avaliação da área valvar mitral por meio da reconstrução multiplano na ecocardiografia tridimensional é restrita a softwares específicos e à experiência dos ecocardiografistas. Eles precisam selecionar manualmente o frame do vídeo que contenha a área de abertura máxima da valva mitral, dimensão fundamental para a identificação de estenose mitral. Objetivo: Automatizar o processo de determinação da área de abertura máxima da valva mitral, por meio da aplicação de Processamento Digital de Imagens (PDI) em exames de ecocardiograma, desenvolvendo um algoritmo aberto com leitura de vídeo no formato avi. Método: Este estudo piloto observacional transversal foi realizado com vinte e cinco exames diferentes de ecocardiograma, sendo quinze com abertura normal e dez com estenose mitral reumática. Todos os exames foram realizados e disponibilizados por dois especialistas, com autorização do Comitê de Ética em Pesquisa, que utilizaram dois modelos de aparelhos ecocardiográficos: Vivid E95 (GE Healthcare) e Epiq 7 (Philips), com sondas multiplanares transesofágicas. Todos os vídeos em formato avi foram submetidos ao PDI através da técnica de segmentação de imagens. Resultados: As medidas obtidas manualmente por ecocardiografistas experientes e os valores calculados pelo sistema desenvolvido foram comparados utilizando o diagrama de Bland-Altman. Observou-se maior concordância entre valores no intervalo de 0,4 a 2,7 cm². Conclusão: Foi possível determinar automaticamente a área de máxima abertura das valvas mitrais, tanto para os casos advindos da GE quanto da Philips, utilizando apenas um vídeo como dado de entrada. O algoritmo demonstrou economizar tempo nas medições quando comparado com a mensuração habitual. (AU)
Background: The evaluation of mitral valve area through multiplanar reconstruction in 3-dimensional echocardiography is restricted to specific software and to the experience of echocardiographers. They need to manually select the video frame that contains the maximum mitral valve opening area, as this dimension is fundamental to identification of mitral stenosis. Objective: To automate the process of determining the maximum mitral valve opening area, through the application of digital image processing (DIP) in echocardiography tests, developing an open algorithm with video reading in avi format. Method: This cross-sectional observational pilot study was conducted with 25 different echocardiography exams, 15 with normal aperture and 10 with rheumatic mitral stenosis. With the authorization of the Research Ethics Committee, all exams were performed and made available by 2 specialists who used 2 models of echocardiographic devices: Vivid E95 (GE Healthcare) and Epiq 7 (Philips), with multiplanar transesophageal probes. All videos in avi format were submitted to DIP using the image segmentation technique. Results: The measurements obtained manually by experienced echocardiographers and the values calculated by the developed system were compared using a Bland-Altman diagram. There was greater agreement between values in the range from 0.4 to 2.7 cm². Conclusion: It was possible to automatically determine the maximum mitral valve opening area, for cases from both GE and Philips, using only 1 video as input data. The algorithm has been demonstrated to save time on measurements when compared to the usual method. (AU)
Assuntos
Humanos , Doenças das Valvas Cardíacas/mortalidade , Valva Mitral/fisiopatologia , Valva Mitral/diagnóstico por imagem , Estenose da Valva Mitral/etiologia , Processamento de Imagem Assistida por Computador/métodos , Doxorrubicina/efeitos da radiação , Ecocardiografia Transesofagiana/métodos , Ecocardiografia Tridimensional/métodos , Substituição da Valva Aórtica Transcateter/métodos , Isoproterenol/efeitos da radiação , Valva Mitral/cirurgiaRESUMO
The kidney is the body's most important organ and the protein components in urine could be detected for diagnosing certain diseases. The amount of IgG protein in urine could be used to determine the degree of kidney function damage. IgG protein in human urine was detected by vertical flow paper-based microfluidic chip, double-antibody sandwich immunoreaction, and cell phone image processing. The results showed that using an IgG antibody concentration of 500 μg/mL and a gold standard antibody concentration of 100 μg/mL, the image signal showed a good linear relationship in the range of IgG concentration of 0.2-3.2 μg/mL, with R2=0.973 3 achieved. A complete set of detection devices were designed and the detection method showed good non-specificity.
Assuntos
Humanos , Microfluídica , Imunoglobulina G , Rim , Técnicas Analíticas MicrofluídicasRESUMO
Objective:To study whether Bayesian penalized likelihood (BPL) and its optimized reconstruction algorithm can improve the reconstructed image quality of low count total-body PET.Methods:Eight patients (5 males, 3 females, age (67.2±6.3) years) who underwent hybrid 18F-FDG PET/MR total-body scans at Department of Nuclear Medicine in Union Hospital, Tongji Medical College, Huazhong University of Science and Technology were collected retrospectively from January to June in 2023. Total-body PET/MR images of them were included and list-mode data were reconstructed with four methods, namely 25% counts conventional reconstruction (group 1), 100% counts conventional reconstruction (group 2), 25% counts BPL reconstruction (group 3), and 25% counts optimized BPL reconstruction (group 4). At last, 32 total-body PET images were obtained. SUV max and SUV mean in different ROIs and tumor metabolic volume (MTV) were measured. Total lesion glycolysis (TLG) and parameters of image quality including the ratio of lesion to background (L/B) and image signal-to-noise ratio (SNR) were calculated. Then the differences in all the parameters among the four groups were analyzed by repeated measures analysis of variance and Friedman test. Quantitative differences between BPL reconstruction and optimized BPL with the 100% counts conventional reconstruction were compared respectively by using the Bland-Altman (BA) plot. Results:For the inter-group comparison, except for SUV mean in the muscle ( F=0.38, P=0.767), SUV max and SUV mean in other ROIs were statistically different ( F values: 8.15-36.08, χ2=18.15, all P<0.01), as well as MTV and L/B ( χ2 values: 10.65, 13.35, P values: 0.014, 0.004), but not for TLG ( χ2=4.95, P=0.175) or SNR ( F=2.64, P=0.076). For the pairwise comparison, the differences between group 2 and group 3 were the most significant (all P<0.05). Compared with group 2, there were no significant differences for SUV max and SUV mean of the cerebellar cortex and lesions in group 4 (all P>0.05), as well as MTV and L/B (both P>0.05). In addition, compared with group 1, SUV max of liver and muscle in group 2 were decreased (both P<0.05), while there were no significant differences in group 4 (all P>0.05). BA plots showed that the differences of SUV, MTV, and TLG between group 4 and group 2 were smaller obviously than those between group 3 and group 2. Conclusion:BPL reconstruction can improve low focus detection sensitivity induced by low counts, but it will cause significant changes for PET quantification, which can be solved by optimized BPL reconstruction.
RESUMO
Purpose/Significance To solve the problems of understanding the connotation of courses for ideological and political edu-cation and how to integrate the ideological and political elements into the courses effectively.Method/Process The study starts from the analysis of the ideological and political education and the characteristics of the digital image processing course in medical universities and colleges,determines curriculum ideological and political education objectives,formulates curriculum ideological and political implementa-tion strategies.Result/Conclusion The study realizes the teaching content design of digital image processing course integrated the ideo-logical and political elements.Through the online and offline combination,the study explores the implementation of ideological and politi-cal thinking in medical universities and colleges courses.