RESUMEN
Several data sets have been collected and various artificial intelligence models have been developed for COVID-19 classification and detection from both chest radiography (CXR) and thorax computed tomography (CTX) images. However, the pitfalls and shortcomings of these systems significantly limit their clinical use. In this respect, improving the weaknesses of advanced models can be very effective besides developing new ones. The inability to diagnose ground-glass opacities by conventional CXR has limited the use of this modality in the diagnostic work-up of COVID-19. In our study, we investigated whether we could increase the diagnostic efficiency by collecting a novel CXR data set, which contains pneumonic regions that are not visible to the experts and can only be annotated under CTX guidance. We develop an ensemble methodology of well-established deep CXR models for this new data set and develop a machine learning-based non-maximum suppression strategy to boost the performance for challenging CXR images. CTX and CXR images of 379 patients who applied to our hospital with suspected COVID-19 were evaluated with consensus by seven radiologists. Among these, CXR images of 161 patients who also have had a CTX examination on the same day or until the day before or after and whose CTX findings are compatible with COVID-19 pneumonia, are selected for annotating. CTX images are arranged in the main section passing through the anterior, middle, and posterior according to the sagittal plane with the reformed maximum intensity projection (MIP) method in the coronal plane. Based on the analysis of coronal MIP reconstructed CTX images, the regions corresponding to the pneumonia foci are annotated manually in CXR images. Radiologically classified posterior to anterior (PA) CXR of 218 patients with negative thorax CTX imaging were classified as COVID-19 pneumonia negative group. Accordingly, we have collected a new data set using anonymized CXR (JPEG) and CT (DICOM) images, where the PA CXRs contain pneumonic regions that are hidden or not easily recognized and annotated under CTX guidance. The reference finding was the presence of pneumonic infiltration consistent with COVID-19 on chest CTX examination. COVID-Net, a specially designed convolutional neural network, was used to detect cases of COVID-19 among CXRs. Diagnostic performances were evaluated by ROC analysis by applying six COVID-Net variants (COVIDNet-CXR3-A, -B, -C/COVIDNet-CXR4-A, -B, -C) to the defined data set and combining these models in various ways via ensemble strategies. Finally, a convex optimization strategy is carried out to find the outperforming weighted ensemble of individual models. The mean age of 161 patients with pneumonia was 49.31 ± 15.12, and the median age was 48 years. The mean age of 218 patients without signs of pneumonia in thorax CTX examination was 40.04 ± 14.46, and the median was 38. When working with different combinations of COVID-Net's six variants, the area under the curve (AUC) using the ensemble COVID-Net CXR 4A-4B-3C was .78, sensitivity 67%, specificity 95%; COVID-Net CXR 4a-3b-3c was .79, sensitivity 69% and specificity 94%. When diverse and complementary COVID-Net models are used together through an ensemble, it has been determined that the AUC values are close to other studies, and the specificity is significantly higher than other studies in the literature.
Asunto(s)
COVID-19 , Radiografía Torácica , SARS-CoV-2 , Tomografía Computarizada por Rayos X , Humanos , COVID-19/diagnóstico por imagen , Tomografía Computarizada por Rayos X/métodos , Radiografía Torácica/métodos , Femenino , Masculino , Aprendizaje Automático , Persona de Mediana Edad , Pulmón/diagnóstico por imagen , Tórax/diagnóstico por imagen , Anciano , Pandemias , Adulto , Neumonía Viral/diagnóstico por imagen , Neumonía Viral/diagnósticoRESUMEN
Objectives: Computed tomography (CT) plays a complementary role in the diagnosis of the pneumonia-burden of COVID-19 disease. However, the low contrast of areas of inflammation on CT images, areas of infection are difficult to identify. The purpose of this study is to develop a post-image-processing method for quantitative analysis of COVID-19 pneumonia-related changes in CT attenuation values using a pixel-based analysis rather than more commonly used clustered focal pneumonia volumes. The COVID-19 pneumonia burden is determined by experienced radiologists in the clinic. Previous AI software was developed for the measurement of COVID-19 lesions based on the extraction of local pneumonia features. In this respect, changes in the pixel levels beyond the clusters may be overlooked by deep learning algorithms. The proposed technique focuses on the quantitative measurement of COVID-19 related pneumonia over the entire lung in pixel-by-pixel fashion rather than only clustered focal pneumonia volumes. Material and Methods: Fifty COVID-19 and 50 age-matched negative control patients were analyzed using the proposed technique and commercially available artificial intelligence (AI) software. The %pneumonia was calculated using the relative volume of parenchymal pixels within an empirically defined CT density range, excluding pulmonary airways, vessels, and fissures. One-way ANOVA analysis was used to investigate the statistical difference between lobar and whole lung %pneumonia in the negative control and COVID-19 cohorts. Results: The threshold of high-and-low CT attenuation values related to pneumonia caused by COVID-19 were found to be between â642.4 HU and 143 HU. The %pneumonia of the whole lung, left upper, and lower lobes were 8.1 ± 4.4%, 6.1 ± 4.5, and 11.3 ± 7.3% for the COVID-19 cohort, respectively, and statistically different (P < 0.01). Additionally, the pixel-based methods correlate well with existing AI methods and are approximately four times more sensitive to pneumonia particularly at the upper lobes compared with commercial software in COVID-19 patients (P < 0.01). Conclusion: Pixel-by-pixel analysis can accurately assess pneumonia in COVID-19 patients with CT. Pixel-based techniques produce more sensitive results than AI techniques. Using the proposed novel technique, %pneumonia could be quantitatively calculated not only in the clusters but also in the whole lung with an improved sensitivity by a factor of four compared to AI-based analysis. More significantly, pixel-by-pixel analysis was more sensitive to the upper lobe pneumonia, while AI-based analysis overlooked the upper lung pneumonia region. In the future, this technique can be used to investigate the efficiency of vaccines and drugs and post COVID-19 effects.
RESUMEN
PURPOSE: To measure the lesion size reduction in eyes with active toxoplasma retinochoroiditis during the disease course with swept-source optical coherence tomography angiography (SS-OCTA). METHODS: We retrospectively analysed the chorioretinal lesion size in a group of 14 eyes with a single active toxoplasma retinochoroiditis lesion. SS-OCTA was performed at the baseline and follow-up in all eyes. The 6 × 6 mm choriocapillaris slab images were evaluated with image analysis (MATLAB). The number of black and white pixels in a 1500-µm-diameter circle centred on each active lesion was counted at the time of baseline examination and at the first follow-up visit when the chorioretinal scar formation was noticed. RESULTS: Fourteen eyes with a single active toxoplasmosis retinochoroiditis lesion were included. Ten patients were female and three were male. The mean age was 29.1 ± 14.9 years. Active lesions were at the macula in five eyes, at the periphery in six eyes and juxtapapillary in three eyes. At the initial examination, the lesion area was observed as an area with a decreased flow signal on SS-OCTA. There was the perilesional capillary disruption in superficial and deep capillary plexi together with a diffuse capillary network attenuation and non-detectable flow signal zones in the choriocapillaris slabs. In addition to sulfamethoxazole-trimethoprim and azithromycin combination, oral corticosteroids were only co-administered in five (35%) eyes with macular involvement. The chorioretinal scar formation was observed in 4 to 16 weeks. At the time of inactivity, the original lesion was diminished in size when compared to its baseline in all study eyes (p = 0.001) with a mean black pixel reduction percentage of 21.8%. The reduction was 15.4% in eyes with macular lesion, 31.6% with peripheral lesions and 18.1% with juxtapapillary lesions (p = 0.001, p = 0.032, p = 0.028, p = 0.043, respectively). Visual acuity was correlated with black pixel reduction percentage in eyes with macular lesion (r = 0.56, p < 0.001). CONCLUSION: Healing of the active toxoplasma retinochoroiditis lesion size could be monitored with an OCTA-based image analysis technique. Interestingly, the reduction in the lesion size was lesser in the macular lesions than the peripheral and juxtapapillary lesions following the treatment and this might contribute to the poorer visual outcomes observed in eyes with macular lesions.