Despite the overall success of using artificial intelligence (AI) to assist radiologists in performing computer-aided patient diagnosis, it remains challenging to build good models with small ...datasets at individual sites. Because many medical images do not come with proper labelling for training, this requires radiologists to perform strenuous labelling work and to prepare the dataset for training. Placing such demands on radiologists is unsustainable, given the ever-increasing number of medical images taken each year. We propose an alternative solution using a relatively new learning framework. This framework, called federated learning, allows individual sites to train a global model in a collaborative effort. Federated learning involves aggregating training results from multiple sites to create a global model without directly sharing datasets. This ensures that patient privacy is maintained across sites. Furthermore, the added supervision obtained from the results of partnering sites improves the global model's overall detection abilities. This alleviates the issue of insufficient supervision when training AI models with small datasets. Lastly, we also address the major challenges of adopting federated learning.
Previous research has demonstrated a correlation between hand grip strength (HGS) and muscle strength. This study aims to determine the relationship between HGS and muscle mass in older Asian adults.
...We retrospectively reviewed the dual-energy X-ray absorptiometry (DXA) records of 907 older adults (239 (26.4%) men and 668 (73.6%) women) at one medical institution in Taipei, Taiwan, from January 2019, to December 2020. Average age was 74.80 ± 9.43 and 72.93 ± 9.09 for the males and females respectively. The inclusion criteria were: 1) aged 60 and older, 2) underwent a full-body DXA scan, and 3) performed hand grip measurements. Patients with duplicate results, incomplete records, stroke history, and other neurological diseases were excluded. Regional skeletal muscle mass was measured using DXA. HGS was measured using a Jamar handheld dynamometer.
Total lean muscle mass (kg) averaged 43.63 ± 5.81 and 33.16 ± 4.32 for the males and females respectively. Average HGS (kg) was 28.81 ± 9.87 and 19.19 ± 6.17 for the males and females respectively. In both sexes, HGS and regional muscle mass consistently declined after 60 years of age. The rates of decline per decade in upper and lower extremity muscle mass and HGS were 7.06, 4.95, and 12.30%, respectively, for the males, and 3.36, 4.44, and 12.48%, respectively, for the females. In men, HGS significantly correlated with upper (r = 0.576, p < 0.001) and lower extremity muscle mass (r = 0.532, p < 0.001). In women, the correlations between HGS and upper extremity muscle mass (r = 0.262, p < 0.001) and lower extremity muscle mass (r = 0.364, p < 0.001) were less strong, though also statistically significant.
Muscle mass and HGS decline with advancing age in both sexes, though the correlation is stronger in men. HGS measurements are an accurate proxy for muscle mass in older Asian adults, particularly in males.
The morphology and distribution of microcalcifications are the most important descriptors for radiologists to diagnose breast cancer based on mammograms. However, it is very challenging and ...time-consuming for radiologists to characterize these descriptors manually, and there also lacks of effective and automatic solutions for this problem. We observed that the distribution and morphology descriptors are determined by the radiologists based on the spatial and visual relationships among calcifications. Thus, we hypothesize that this information can be effectively modelled by learning a relationship-aware representation using graph convolutional networks (GCNs). In this study, we propose a multi-task deep GCN method for automatic characterization of both the morphology and distribution of microcalcifications in mammograms. Our proposed method transforms morphology and distribution characterization into node and graph classification problem and learns the representations concurrently. We trained and validated the proposed method in an in-house dataset and public DDSM dataset with 195 and 583 cases,respectively. The proposed method reaches good and stable results with distribution AUC at 0.812<inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula>0.043 and 0.873<inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula>0.019, morphology AUC at 0.663<inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula>0.016 and 0.700<inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula>0.044 for both in-house and public datasets. In both datasets, our proposed method demonstrates statistically significant improvements compared to the baseline models. The performance improvements brought by our proposed multi-task mechanism can be attributed to the association between the distribution and morphology of calcifications in mammograms, which is interpretable using graphical visualizations and consistent with the definitions of descriptors in the standard BI-RADS guideline. In short, we explore, for the first time, the application of GCNs in microcalcification characterization that suggests the potential of using graph learning for more robust understanding of medical images.
The benefits of mammographic screening have been shown to include a decrease in mortality due to breast cancer. Taiwan's Breast Cancer Screening Program is a national screening program that has ...offered biennial mammographic breast cancer screening for women aged 50-69 years since 2004 and for those aged 45-69 years since 2009, with the implementation of mobile units in 2010. The purpose of this study was to compare the performance results of the program with changes in the previous (2004-2009) and latter (2010-2020) periods.
A cohort of 3,665,078 women who underwent biennial breast cancer mammography screenings from 2004 to 2020 was conducted, and data were obtained from the Health Promotion Administration, Ministry of Health and Welfare of Taiwan. We compared the participation of screened women and survival rates from breast cancer in the earlier and latter periods across national breast cancer screening programs.
Among 3,665,078 women who underwent 8,169,869 examinations in the study population, the screened population increased from 3.9% in 2004 to 40% in 2019. The mean cancer detection rate was 4.76 and 4.08 cancers per 1000 screening mammograms in the earlier (2004-2009) and latter (2010-2020) periods, respectively. The 10-year survival rate increased from 89.68% in the early period to 97.33% in the latter period. The mean recall rate was 9.90% (95% CI: 9.83-9.97%) in the early period and decreased to 8.15% (95%CI, 8.13-8.17%) in the latter period.
The evolution of breast cancer screening in Taiwan has yielded favorable outcomes by increasing the screening population, increasing the 10-year survival rate, and reducing the recall rate through the participation of young women, the implementation of a mobile unit service and quality assurance program, thereby providing historical evidence to policy makers to plan future needs.
Given that early identification of breast cancer type allows for less-invasive therapies, we aimed to develop a machine learning model to discriminate between ductal carcinoma in situ (DCIS) and ...minimally invasive breast cancer (MIBC).
In this retrospective study, the health records of 420 women who underwent biopsies between 2010 and 2020 to confirm breast cancer were collected. A trained XGBoost algorithm was used to classify cancers as either DCIS or MIBC using clinical characteristics, mammographic findings, ultrasonographic findings, and histopathological features. Its performance was measured against other methods using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, precision, and F1 score.
The model was trained using 357 women and tested using 63 women with an overall 420 patients (mean standard deviation age, 57.1 12.0 years). The model performed well when feature importance was determined, reaching an accuracy of 0.84 (95% confidence interval CI, 0.76-0.91), an AUC of 0.93 (95% CI, 0.87-0.95), a specificity of 0.75 (95% CI, 0.67-0.83), and a sensitivity of 0.91 (95% CI, 0.76-0.94).
The XGBoost model, combining clinical, mammographic, ultrasonographic, and histopathologic findings, can be used to discriminate DCIS from MIBC with an accuracy equivalent to that of experienced radiologists, thereby giving patients the widest range of therapeutic options.
The purpose of this study was to investigate the influence of arterial input function (AIF) selection on the quantification of vertebral perfusion using axial dynamic contrast-enhanced magnetic ...resonance imaging (DCE-MRI). In this study, axial DCE-MRI was performed on 2 vertebrae in each of eight healthy volunteers (mean age, 36.9 years; 5 men) using a 1.5-T scanner. The pharmacokinetic parameters K
, v
, and v
, derived using a Tofts model on axial DCE-MRI of the lumbar vertebrae, were evaluated using various AIFs: the population-based aortic AIF (AIF_PA), a patient-specific aortic AIF (AIF_A) and a patient-specific segmental arterial AIF (AIF_SA). Additionally, peaks and delay times were changed to simulate the effects of various AIFs on the calculation of perfusion parameters. Nonparametric analyses including the Wilcoxon signed rank test and the Kruskal-Wallis test with a Dunn-Bonferroni post hoc analysis were performed. In simulation, K
and v
increased as the peak in the AIF decreased, but v
increased when delay time in the AIF increased. In humans, the estimated K
and v
were significantly smaller using AIF_A compared to AIF_SA no matter the computation style (pixel-wise or region-of-interest based). Both these perfusion parameters were significantly greater using AIF_SA compared to AIF_A.
Radiosensitivity in the breasts increases the risk of carcinogenesis from exposure to the ionizing radiation of computed tomography (CT) administered in the course of medical attention. Bismuth ...shielding techniques have been used to reduce radiation, but image noise increased, degrading image quality.
The aim of this study was to investigate how the use of iterative reconstruction (IR) combined with bismuth shielding influences image quality.
Women aged at least 20 years with body mass indexes <28 were recruited and randomly assigned to 1 of 3 CT scanning protocols without shielding, with a bismuth breast shield before the scout view, or with a bismuth breast shield after the scout view. All obtained images were reconstructed using an IR algorithm. To evaluate radiation dose, 2 Gafchromic films were placed over the clothes, 1 near each nipple.
Average dose reduction was significant (27.99%, P < .05) when bismuth shielding was applied after the scout view. Using the contrast-to-noise ratio, the image quality was found to be superior when the IR algorithm was applied. Using quantitative evaluations by 2 radiologists applying a 4-point Likert scale, significant differences in image quality were not found among the 3 protocols.
Bismuth breast shields, particularly when used after acquiring scout images, are effective at reducing radiation dose without undermining the diagnostic value of the images when the IR technique is applied.
The Breast Imaging-Reporting and Data System (BI-RADS) atlas defines category 5 assessments as appropriate only for lesions that are almost certainly cancer, with a positive predictive value (PPV) of ...≥95%. This study aims to demonstrate the feasibility of classifying lesions at diagnostic breast imaging with sufficiently high PPV to merit category 5 assessments, and to identify those lesion descriptors that yield such a high PPV.
For this Health Insurance Portability and Accountability Act compliant and IRB exempt study, we reviewed diagnostic breast imaging examinations (mammography and/or ultrasound) assessed as highly suggestive of malignancy (BI-RADS category 5). Pathology diagnosis was considered the gold standard. PPV
(biopsy performed) was calculated, and the BI-RADS descriptors for each lesion were analyzed.
Among 22,564 consecutive diagnostic breast imaging examinations between January 2010 and September 2015, we identified 239 exams (1.1%) assessed as BI-RADS category 5 (mean age 62.5 years). Malignancy (invasive breast carcinoma and/or ductal carcinoma in situ) was diagnosed in 233 examinations (PPV
97.5% and 95% confidence interval: 96.2%-98.8%). The most common lesion types were mass (170) and calcifications (116). Of the 220 examinations involving both mammography and ultrasound, no category 5 lesions had <3 suspicious BI-RADS descriptors, only three lesions had three suspicious descriptors, but the remaining 217 lesions (98.6%) had ≥4 suspicious descriptors.
In clinical practice, it is feasible to make BI-RADS category 5 assessments with the intended ≥95% PPV. To justify a category 5 assessment, at least four suspicious BI-RADS descriptors should be identified at the combination of diagnostic mammography and ultrasound examinations.
Purpose: We aimed to develop a novel interpretable artificial intelligence (AI) model algorithm focusing on automatic detection and classification of various patterns of calcification distribution in ...mammographic images using a unique graph convolution approach. Materials and methods: Images from 292 patients, which showed calcifications according to the mammographic reports and diagnosed breast cancers, were collected. The calcification distributions were classified as diffuse, segmental, regional, grouped, or linear. Excluded were mammograms with (1) breast cancer with multiple lexicons such as mass, asymmetry, or architectural distortion without calcifications; (2) hidden calcifications that were difficult to mark; or (3) incomplete medical records. Results: A graph-convolutional-network-based model was developed. A total of 581 mammographic images from 292 cases of breast cancer were divided based on the calcification distribution pattern: diffuse (n = 67), regional (n = 115), group (n = 337), linear (n = 8), or segmental (n = 54). The classification performances were measured using metrics including precision, recall, F1 score, accuracy, and multi-class area under the receiver operating characteristic curve. The proposed model achieved a precision of 0.522 ± 0.028, sensitivity of 0.643 ± 0.017, specificity of 0.847 ± 0.009, F1 score of 0.559 ± 0.018, accuracy of 64.325 ± 1.694%, and area under the curve of 0.745 ± 0.030; thus, the method was found to be superior compared to all baseline models. The predicted linear and diffuse classifications were highly similar to the ground truth, and the predicted grouped and regional classifications were also superior compared to baseline models. The prediction results are interpretable using visualization methods to highlight the important calcification nodes in graphs. Conclusions: The proposed deep neural network framework is an AI solution that automatically detects and classifies calcification distribution patterns on mammographic images highly suspected of showing breast cancers. Further study of the AI model in an actual clinical setting and additional data collection will improve its performance.