Ken Enda’s research while affiliated with Hokkaido University and other places

Contrast-enhanced ultrasound (CEUS) plays a pivotal role in the diagnosis of primary breast cancer and in axillary lymph node (ALN) metastasis. However, the imaging features that are clinically crucial for lymph node metastasis have not been fully elucidated. Hence, we developed a bimodal model to predict ALN metastasis in patients with early breast cancer by integrating CEUS images with the annotated imaging features. The model adopted a light-gradient boosting machine to produce feature importance, enabling the extraction of clinically crucial imaging features. In this retrospective study, the diagnostic performance of the model was investigated using 788 CEUS images of ALNs obtained from 788 patients who underwent breast surgery between 2013 and 2021, with the ground truth defined by the pathological diagnosis. The results indicated that the test cohort had an area under the receiver operating characteristic curve (AUC) value of 0.93 (95% confidence interval: 0.88, 0.98). The model had an accuracy of 0.93, which was higher than the radiologist’s diagnosis (accuracy of 0.85). The most important imaging features were heterogeneous enhancement, diffuse cortical thickening, and eccentric cortical thickening. Our model has an excellent diagnostic performance, and the extracted imaging features could be crucial for confirming ALN metastasis in clinical settings.

Background: Developmental dysplasia of the hip (DDH) is a prevalent issue in infants, with ultrasound crucial for early detection. Existing automatic diagnostic models lack precision due to noise, but 3D technology may enhance it. This study aimed to create and assess a deep-learning-based model for automated DDH diagnosis by using 3D transformation technology on two-dimensional ultrasound images. Methods: A retrospective study of 417 infants at risk of DDH used ultrasound images, combining convolutional neural networks and image processing. The images were analyzed using algorithms such as HigherHRNet-W48. The approach included apex point estimation, signal heterogeneity analysis of ilium, which focused on the bony area with high intensity and evaluate ilium rotation, alpha angle creation, and the establishment of a comprehensive method for DDH diagnosis. Results: Key findings include: (1) Superior accuracy in apex point estimation by the HigherHRNet-W48 model, even better than orthopedic residents. (2) Thorough quality assessments of ultrasound images, leading to qualified and disqualified categories, with qualified images displaying notably lower error rates. (3) The AUC of the model for DDH detection in the qualifying images was 0.92, exceeding the diagnostic accuracy of the resident, indicating the diagnostic capability of the tool. Conclusions: The study developed a deep-learning-based model for DDH detection in infants, melding 3D technology with deep learning to address challenges like noise and rotation in ultrasound images. The study’s innovation demonstrated a comparative accuracy to specialized evaluations, even with non-specialist images, highlighting its potential to assist novice sonographers and enhance diagnostic precision.

Foundation models pretrained on large-scale pathology datasets have shown promising results across various diagnostic tasks. Here, we present a systematic evaluation of transfer learning strategies for brain tumor classification using these models. We analyzed 252 cases comprising five major tumor types: glioblastoma, astrocytoma, oligodendroglioma, primary central nervous system lymphoma, and metastatic tumors. Comparing state-of-the-art foundation models with conventional approaches, we found that foundation models demonstrated robust classification performance with as few as 10 patches per case, challenging the traditional assumption that extensive per-case image sampling is necessary. Furthermore, our evaluation revealed that simple transfer learning strategies like linear probing were sufficient, while fine-tuning often degraded model performance. These findings suggest a paradigm shift from extensive data collection to efficient utilization of pretrained features, providing practical implications for implementing AI-assisted diagnosis in clinical pathology.

Bimodal convolutional neural networks (CNNs) are frequently combined with patient information or several medical images to enhance the diagnostic performance. However, the technologies that integrate automatically generated clinical measurements within the images are scarce. Hence, we developed a bimodal model that produced automatic algorithm for clinical measurement (aaCM) from radiographic images and integrated the model with CNNs. In this multicenter research project, the diagnostic performance of the model was investigated with 813 radiographic hip images of infants at risk of developmental dysplasia of the hips (232 and 581 images of unstable and stable hips, respectively), with the ground truth defined by provocative examinations. The results indicated that the accuracy of aaCM was equal or higher than that of specialists, and the bimodal model showed better diagnostic performance than LightGBM, XGBoost, SVM, and single CNN models. aaCM can provide expert’s knowledge in a high level, and our proposed bimodal model has better performance than the state-of-art models.

Considering explainability is crucial in medical artificial intelligence, technologies to quantify Grad-CAM heatmaps and perform automatic integration based on domain knowledge remain lacking. Hence, we created an end-to-end model that produced CAM scores on regions of interest (CSoR), a measure of relative CAM activity, and feature importance scores by automatic algorithms for clinical measurement (aaCM) followed by LightGBM. In this multicenter research project, the diagnostic performance of the model was investigated with 813 radiographic hip images in infants at risk of unstable hips, with the ground truth defined by provocative examinations. The results indicated that the accuracy of aaCM was higher than that of specialists, and the model with ad hoc adoption of aaCM outperformed the image-only-based model. Subgroup analyses in positive cases indicated significant differences in CSoR between the unstable and contralateral sides despite containing only binary labels (positive or negative). In conclusion, aaCM reinforces the performance, and CSoR potentially indicates model reliability.

We report a fatal subdural empyema caused by Campylobacter rectus in a 66-year-old female who developed acute onset of confusion, dysarthria, and paresis in her left extremities. A CT scan showed hypodensity in a crescentic formation with a mild mid-line shift. She had a bruise on her forehead caused by a fall several days before admission, which initially raised subdural hematoma (SDH) diagnosis, and a burr hole procedure was planned. However, her condition deteriorated on the admission night, and she died before dawn. An autopsy revealed that she had subdural empyema (SDE) caused by Campylobacter rectus and Slackia exigua. Both microorganisms are oral microorganisms that rarely cause extra-oral infection. In our case, head trauma caused a skull bone fracture, and sinus infection might have expanded to the subdural space causing SDE. CT/MRI findings were not typical for either SDH or SDE. Early recognition of subdural empyema and prompt initiation of treatment with antibiotics and surgical drainage is essential for cases of SDE. We present our case and a review of four reported cases. Keywords Campylobacter rectus; Empyema, Subdural; Hematoma, Subdural; Sinusitis

Background: The number of patients with fragility fracture has been increasing. Although the increasing number of patients with fragility fracture increased the rate of fracture (refracture), the causes of refracture are multifactorial, and its predictors are still not clarified. In this issue, we collected a registry-based longitudinal dataset that contained more than 7000 patients with fragility fractures treated surgically to detect potential predictors for clinical refracture. Methods: Based on the fact that machine learning algorithms are often used for the analysis of a large-scale dataset, we developed automatic prediction models and clarified the relevant features for patients with clinical refracture. Formats of input data containing perioperative clinical information were table data. Clinical refracture was documented as the primary outcome if the diagnosis of fracture was made at postoperative outpatient care. A decision-tree-based model, LightGBM, had moderate accuracy for the prediction in the test and the independent dataset, whereas the other models had poor accuracy or worse. Results: From a clinical perspective, rheumatoid arthritis (RA) and chronic kidney disease (CKD) were noted as the relevant features for patients with clinical refracture, both of which were associated with secondary osteoporosis. Conclusion: The decision-tree-based algorithm showed the precise prediction of clinical refracture, in which RA and CKD were detected as the potential predictors. Understanding these predictors may improve the management of patients with fragility fractures.