Hsian-He Hsu

Background/Objectives: To develop and externally evaluate a deep learning framework for multi-label thoracic disease classification on chest radiographs using hybrid convolutional neural network (CNN)–transformer architectures, hierarchical scalar-weighted fusion, and ensemble strategies. Methods: This retrospective, multi-center study utilized publicly available datasets: NIH ChestX-ray14 (112,120 images; 30,805 patients) for model development and internal testing, and CheXpert (223,415 images) plus ChestX-Det10 (3578 images) for external validation. Nine CNN–transformer hybrids were systematically benchmarked, and the proposed model incorporated multi-scale DenseNet121 features, scalar-weighted fusion, positional encodings, and cross-attention. Four post hoc ensemble methods were explored, including a class-wise Top-3 Grid Search. Performance was evaluated using AUROC as the primary metric, along with precision, recall, F1-score, accuracy, specificity, positive predictive value, and negative predictive value. Statistical comparisons were performed using bootstrapped resampling and appropriate parametric or non-parametic tests. Results: On the NIH internal test set, the proposed hybrid model achieved a mean AUROC of 0.8495, which was significantly higher than that of the DenseNet121 baseline (0.8441, p = 0.032). The Top-3 Grid Search ensemble further improved internal performance, achieving a mean AUROC of 0.8577 (p < 0.00001). On external validation, the ensemble consistently outperformed DenseNet121, achieving mean AUROCs of 0.6500 on CheXpert (p < 0.001) and 0.6592 on ChestX-Det10 (p < 0.001). Per-class analysis revealed significant improvements for clinically important conditions such as cardiomegaly, mass, and pneumothorax. Grad-CAM visualizations demonstrated the strong alignment of predicted abnormalities with radiologically relevant regions. Conclusions: This CNN–transformer framework, particularly when combined with class-wise ensemble strategies, provided modest but statistically significant improvements in multi-label chest X-ray classification. External validation suggested partial generalizability across datasets, although performance remained moderate under domain shift.

Background and Objectives: The needle path is crucial for preoperative localization of deep thoracic pulmonary nodules using CT-guided patent blue dye (PBD) injection. This study aimed to evaluate the efficacy and safety of four categorized needle approach paths tailored to the anatomical location of the nodules. Materials and Methods: We retrospectively evaluated data from 50 consecutive patients (50 deep pulmonary nodules), who underwent CT-guided localization with PBD injection, between November 2015 and May 2023 at our hospital. The nodules could be divided into four categories: (1) perifissural nodules, (2) paravertebral nodules, (3) paramediastinal nodules, and (4) deep parenchymal nodules, according to their location relative to the thoracic organs and the visceral pleura. Needle approach methods and needle pathway lengths were recorded. Clinical and radiological features, technical information, pathological results, and procedure-related complications were analyzed. Results: All 50 dyes were successfully identified by thoracoscopy and then resected without major complication. The mean nodule diameter and the nodular depth were 10.3 (range, 4.7–21.0) mm and 16.1 (range, 0.1–52.2) mm. The needle pathway length was 7.7 (range, 4.5–11.7) cm. The mean procedure time was 16 (range, 8–26) minutes. Asymptomatic pneumothorax developed in twenty-four patients (48.0%), and focal parenchymal hemorrhage occurred in four patients (8.0%) after localization. No patients required chest tube insertion or resuscitation. Conclusions: Strategic needle approach paths provide precise localization of deep thoracic pulmonary nodules with minimal complications. These methods offer a practical framework for improving thoracoscopic surgery in challenging deep thoracic cases.

Background and Objectives: The needle path is crucial for preoperative localization of deep thoracic pulmonary nodules using CT-guided patent blue dye (PBD) injection. This study aimed to evaluate the efficacy and safety of four categorized needle approach paths tailored to the anatomical location of the nodules. Materials and Methods: We retrospectively evaluated data from 50 consecutive patients (50 deep pulmonary nodules), who underwent CT-guided localization with PBD injection, between November 2015 and May 2023 at our hospital. The nodules could be divided into four categories: (1) perifissural nodules (2) paravertebral nodules, (3) paramediastinal nodules, and (4) deep parenchymal nodules, according to the location relative to the thoracic organs and the visceral pleura. The needle approach methods and the needle pathway lengths were recorded. Clinical and radiological features, technical information, pathological results and procedure-related complications were analyzed. Results: All 50 dyes were successfully identified by thoracoscopy and then resected without major complication. The mean nodule diameter and the nodular depth were 10.3 (range, 4.7–21.0) mm, and 16.1 (range, 0.1–52.2) mm. The needle pathway length was 7.7 (range, 4.5–11.7) cm. The mean procedure time was 16 (range, 8–26) minutes. Asymptomatic pneumothorax developed in 24 patient (48.0%), and focal parenchymal hemorrhage occurred in 4 patients (8.0%) after localization. No patients required chest tube insertion or resuscitation. Conclusions: Strategic needle approach paths provide precise localization of deep thoracic pulmonary nodules with minimal complications. These methods offer a practical framework for improving thoracoscopic surgery in challenging deep thoracic cases.

Background and Objectives: Clinically, a major challenge of multiple nodule localization is puncture-related pneumothorax, which may hamper the successful localization. This study aims to investigate and compare the efficacy and safety of the simultaneous and sequential patent blue dye (PBD) injections for identifying multiple pulmonary nodules during preoperative CT-guided localization. Materials and Methods: Sixty-one consecutive patients with multiple pulmonary nodules who underwent preoperative CT-guided localization with PBD injections between January 2020 and December 2020 were retrospectively enrolled. Of these patients, 31 patients with 64 nodules who underwent simultaneous injections were designated as the simultaneous group; the remaining 30 patients with 63 nodules who underwent sequential punctures were designated as the sequential group. The clinical and radiological features, technical information, pathological results, and procedure-related variables and complications of the two groups were reviewed and analyzed. Results: The localization success rate of the simultaneous group was higher than that of the sequential group (100% [64/64] vs. 93.7% [59/63], p = 0.041). The incidences of pneumothorax (32.3 vs. 33.3%, p = 0.929) and pulmonary hemorrhage (6.3 vs. 3.0%, p = 1) were not significantly different between the two groups, and all cases were minor, which did not require further intervention. Additionally, a significantly lower radiation dose (2.7 vs. 3.5 mSv, p = 0.001) and a shorter procedure time (20.95 vs. 25.28 min, p = 0.001) were observed in the simultaneous group than in the sequential group. Conclusions: Compared with the sequential method, simultaneous PBD injections may improve the localization success rate with a shorter procedure time and less radiation exposure if the patient with multiple pulmonary nodules can be approached in a single position. Further prospective studies are needed to validate these results.

Objectives To investigate thin-section computed tomography (CT) features of pulmonary subsolid nodules (SSNs) with sizes between 5 and 20 mm to determine predictive factors for differentiating focal interstitial fibrosis (FIF) from adenocarcinoma. Methods From January 2017 to December 2018, 169 patients who had persistent SSNs 5–20 mm in size and underwent preoperative nodule localization were enrolled. Patient characteristics and thin-section CT features of the SSNs were reviewed and compared between the FIF and adenocarcinoma groups. Univariable and multivariable analyses were used to identify predictive factors of malignancy. Receiver operating characteristic (ROC) curve analysis was used to quantify the performance of these factors. Results Among the 169 enrolled SSNs, 103 nodules (60.9%) presented as pure ground-glass opacities (GGOs), and 40 (23.7%) were FIFs. Between the FIF and adenocarcinoma groups, there were significant differences (p< 0.05) in nodule border, shape, thickness, and coronal/axial (C/A) ratio. Multivariable analysis demonstrated that a well-defined border, a nodule thickness >4.2, and a C/A ratio >0.62 were significant independent predictors of malignancy. The performance of a model that incorporated these three predictors in discriminating FIF from adenocarcinoma achieved a high area under the ROC curve (AUC, 0.979) and specificity (97.5%). Conclusions For evaluating persistent SSNs 5–20 mm in size, the combination of a well-defined border, a nodule thickness > 4.2, and a C/A ratio > 0.62 is strongly correlated with malignancy. High accuracy and specificity can be achieved by using this predictive model. Key Points • Thin-section coronal images play an important role in differentiating FIF from adenocarcinoma. • The combination of a well-defined border, nodule thickness>4.2 mm, and C/A ratio >0.62 is associated with malignancy. • This predictive model may be helpful for managing persistent SSNs between 5 and 20 mm in size.

Objective: To understand the status of pre-procedural safety practices in radiological examinations at radiology residency training institutions in various Asian regions. Methods: A questionnaire based on the Joint Commission International Accreditation Standards was electronically sent to 3 institutions each in 10 geographical regions across 9 Asian countries. Questions addressing 45 practices were divided into 3 categories. A five-tier scale with numerical scores was used to evaluate safety practices in each institution. Responses obtained from three institutions in the United States were used to validate the execution rate of each surveyed safety practice. Results: The institutional response rate was 70.0% (7 Asian regions, 21 institutions). 44 practices (all those surveyed except for the application of wrist tags for identifying patients with fall risks) were validated using the US participants. Overall, the Asian participants reached a consensus on 89% of the safety practices. Comparatively, most Asian participants did not routinely perform three pre-procedural practices in the examination appropriateness topic. Conclusion: Based on the responses from 21 participating Asian institutions, most routinely perform standard practices during radiological examinations except when it comes to examination appropriateness. This study can provide direction for safety policymakers scrutinizing and improving regional standards of care. Advances in knowledge: This is the first multicenter survey study to elucidate pre-procedural safety practices in radiological examinations in seven Asian regions.

Aortic calcification (AC) is associated with increased risks of cardiovascular events and mortality. Numerous studies have explored the association between calcification and abdominal artery aneurysm. However, evidence regarding the association between AC and acute aortic dissection (AAD) is limited. We aimed to evaluate the association between AC-related variables and the development of intimal tear (IT) in patients with AAD. We conducted a retrospective observational study involving 64 patients with type A AAD and 32 patients with type B AAD from February, 2011 to January, 2017 at a tertiary referral medical center in Taiwan. We used the default analysis module “calcification score analysis” to calculate all the calcification variables, including AC scores and volume. We identified an association between AC and AAD. Patients with AAD had a greater AC volume in the aortic arch and greater AC scores for both the ascending aorta and the aortic arch than did patients without AAD. However, hypertension and coronary artery disease, rather than AC remained to be the independent risk factor for AAD in multivariate analysis. Patients with type A AAD had greater mean and cumulative AC volumes in the aortic arch, greater cumulative AC volumes in the whole aorta and higher cumulative AC scores in the aortic arch than did patients with type B AAD. ACs were superimposed on ITs in nearly half of the patients with AAD. In patients with type A AAD, AC was more commonly located distal to the IT and farther from the IT. We identified the associations between AC-related variables and the location of IT in patients with AAD. However, AC was not an independent risk factor for AAD. The distribution of AC was different between patients with type A and type B AAD.