今泉 和良

In lung cytology, screeners and pathologists examine many cells in cytological specimens and describe their corresponding imaging findings. To support this process, our previous study proposed an image-finding generation model based on convolutional neural networks and a transformer architecture. However, further improvements are required to enhance the accuracy of these findings. In this study, we developed a cytology-specific image-finding generation model using a vision transformer and open-source large language models. In the proposed method, a vision transformer pretrained on large-scale image datasets and multiple open-source large language models was introduced and connected through an original projection layer. Experimental validation using 1059 cytological images demonstrated that the proposed model achieved favorable scores on language-based evaluation metrics and good classification performance when cells were classified based on the generated findings. These results indicate that a task-specific model is an effective approach for generating imaging findings in lung cytology.

In the diagnosis of lung cancer, imaging findings of lung nodules are essential for benign and malignant classifications. Although numerous studies have investigated the classification of lung nodules, no method has been proposed for obtaining detailed imaging findings. This study aimed to develop a novel method for generating image findings and classifying benign and malignant nodules in chest computed tomography (CT) images using vision–language models. In this study, we collected chest CT images of 77 patients diagnosed with either benign or malignant tumors at Fujita Health University Hospital. For these images, we cropped the regions of interest around the nodules, and a pulmonologist provided the corresponding image findings. We used vision–language models for image captioning to generate image findings. The findings generated by these two models were grammatically correct, with no deviations in notation, as expected from the image findings. Moreover, the descriptions of benign and malignant characteristics were accurately obtained. The bootstrapping language–image pretraining (BLIP) base model achieved an accuracy of 79.2% in classifying nodules, and the bilingual evaluation understudy-4 score for agreement with physician findings was 0.561. These results suggest that the proposed method may be effective for classifying and generating lung nodule findings.

OBJECTIVES: To develop a comprehensive machine learning model incorporating various clinical factors, including frailty and comorbidities, to predict 30-day readmission and mortality risk in patients with chronic obstructive pulmonary disease (COPD). METHODS: This retrospective cohort study used electronic health records (EHR) from Fujita Health University Hospital (2004-2019) for 1294 patients with COPD and 3499 hospitalization or death events. The EHR contained longitudinal patient data (demographics, diagnoses, test results, clinical records). We developed two eXtreme Gradient Boosting models, the comprehensive Top64 and practical 11-feature models. We compared these with the Comorbidity, Obstruction, Dyspnea, and Previous Exacerbations index (CODEX) model, a widely used tool for predicting hospital readmission or death in patients with COPD. The area under the receiver operating characteristic curve (AUC) with 95% confidence interval (CI), sensitivity, and specificity were used to evaluate the model performance. RESULTS: The Top64 (AUC: 0.769, 95% CI: 0.747-0.791) and practical 11-feature (AUC: 0.746, 95% CI: 0.730-0.762) models performed better than the CODEX model (AUC: 0.587, 95% CI: 0.563-0.611). The Top64 model showed 0.978 sensitivity and 0.341 specificity, and the practical 11-feature model achieved 0.955 sensitivity and 0.361 specificity. The calibration curves showed good agreement between the observed and predicted results for both models. CONCLUSIONS: A machine learning approach based on clinical data readily available from the EHR performed better than existing models in predicting 30-day readmission and mortality risks in patients with COPD. A comprehensive risk prediction tool may enhance individualized care strategies and improve patient outcomes in COPD management.

Bronchoscopic lung volume reduction (BLVR) with endobronchial valves is an established treatment for selected patients with advanced emphysema. A 74-year-old male patient with chronic obstructive pulmonary disease and severe dyspnea was scheduled to undergo BLVR targeting the right middle lobe bronchus based on high-resolution CT findings, which showed severe emphysematous changes with hyperinflation and fissure completeness of 98% in the right middle lobe. The physician conducted preoperative virtual reality (VR)-assisted planning using the patient's imaging data, enabling comprehensive visualisation of the bronchial tree, airway measurements, and procedural simulation. The Chartis system confirmed a 'no flow' pattern, supporting the absence of collateral ventilation. During the procedure, a size 5.5 valve was placed in the right B4/5 bronchus following VR and intraoperative assessments. The patient remained stable postoperatively without complications. VR enhanced procedural planning by improving airway assessment, optimising valve sizing, and reducing cognitive load, leading to increased efficiency and operator confidence. Further research is warranted to validate the utility of VR in bronchoscopic interventions.