Suk Hyun Lee

This study explored the feasibility of developing a model that can diagnose positive and negative bone metastasis from bone scan images using Teachable Machine by Google, a no-code AI platform that does not require programming skills or a GPU environment. A fourth-year medical student trained deep learning models using a Teachable Machine on a dataset of 4626 bone scan images from patients with cancer (mean age 65.1 ± 11.3 years; 50.5% female). Because of severe class imbalance (bone metastasis positive:negative = 400:4226), we compared the diagnostic performance of two strategies (original set and augmented dataset with tenfold data augmentation applied to positive images). We investigated the diagnostic performance using various hyperparameters (epochs 50–1000, batch sizes 16–32) with a learning rate of 0.001. The final model generated by Teachable Machine was compared with a conventional deep learning model based on ResNet50. The combination of epoch = 150 and batch size = 16 showed the optimal diagnostic performance. The overall sensitivity, specificity, and positive and negative predictive values were 57.1%, 93.9%, 90.4%, and 68.7%, respectively. Both Teachable Machine and ResNet50 showed good diagnostic performance (area under the curve = 0.812 and 0.869, respectively), although the diagnostic performance of Teachable Machine was inferior to that of the conventional ResNet50 model (p < 0.001). Given its convenience, Teachable Machine represents a valuable and accessible tool for medical education and preliminary model development. It allows researchers without programming skills or GPU resources to construct feasibility models for medical image classification.

Background and Objectives: Dual-phase amyloid PET imaging has been proposed to provide complementary information regarding amyloid burden and cerebral perfusion. This exploratory pilot study evaluated whether semiquantitative parameters derived from dual-phase PET/CT could differentiate individuals operationally classified as Alzheimer’s disease with mild functional impairment (AD-MFI) from those with mild cognitive impairment (MCI). Materials and Methods: Twenty-four participants (AD-MFI, n = 19; MCI, n = 5) underwent dual-phase amyloid PET/CT and structural MRI. Early phase SUV (eSUV), delayed-phase SUV (dSUV), standardized uptake value ratios (SUVR), and the difference between early and delayed uptake (SUV_diff) were analyzed across predefined cortical regions. Group differences were assessed using nonparametric tests, with false discovery rate (FDR) and Bonferroni corrections applied for multiple comparisons. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis. Results: Several regional parameters demonstrated nominally significant group differences in uncorrected analyses; however, none remained statistically significant after correction for multiple comparisons. Among the evaluated metrics, SUV_diff demonstrated the highest diagnostic performance (sensitivity 84.2%, specificity 80.0%), followed by eSUV (68.4%, 100%) and MRI cortical volume (47.4%, 100%). Delayed-phase parameters alone showed limited discriminatory robustness despite observed group-level differences. Conclusions: In this exploratory cohort, SUV_diff showed moderate discriminatory potential between AD-MFI and MCI. However, given the small sample size and multiplicity of comparisons, the results should be interpreted as hypothesis-generating. Larger, prospective studies are required to determine the reproducibility and clinical utility of dual-phase semiquantitative parameters.

Background: Visual assessment of coronary artery calcium (CAC) on ungated chest CT has been described previously. However, its reliability and clinical utility remain uncertain, particularly in PET/CT studies that use low-dose, low-slice CT and are susceptible to respiratory artifacts. Methods: We retrospectively analyzed 106 patients (median age, 66 years [interquartile range, 60–75 years]; 67 men [63.2%]) who underwent PET/CT and electrocardiogram (ECG)-gated chest CT within a 90-day interval. Six readers (three radiologists and three nuclear medicine physicians) independently assessed CAC on PET/CT using a standard four-point visual scale and a 0–12 ordinal scale based solely on written instructions. Agatston scoring was also performed. Interobserver agreement and concordance with ECG-gated chest CT Agatston score categories were calculated. Major adverse cardiovascular events (MACE) were recorded over a median follow-up of 3.5 years. Results: Interobserver agreement was good for both the standard visual (κ = 0.761) and ordinal (κ = 0.779) scales. Concordance with ECG-gated CT Agatston categories was higher for standard visual (κ = 0.849) and ordinal (κ = 0.750) scoring than for PET/CT Agatston categories (κ = 0.464). Both qualitative scales tended to underestimate CAC categories compared with ECG-gated CT; however, severe CAC on PET/CT predicted MACE (hazard ratios: 4.41 standard visual; 6.59 ordinal), and the ordinal scale significantly stratified MACE-free survival (p = 0.047). Conclusions: Standard visual and ordinal CAC scoring on the ungated CT portion of PET/CT is quick, reproducible, closely mirrors ECG-gated-CT Agatston grading, and offers prognostic value for future MACE in cancer patients.

Purpose: AI-driven scan time reduction is rapidly transforming medical imaging with benefits such as improved patient comfort and enhanced efficiency. A Dual Contrastive Learning Generative Adversarial Network (DCLGAN) was developed to predict full-time PET scans from shorter, noisier scans, improving challenges in imaging patients with movement disorders. Patients and Methods: ¹⁸F FP-CIT PET/CT data from 391 patients with suspected Parkinsonism were used [250 training/validation, 141 testing (hospital A)]. Ground truth (GT) images were reconstructed from 15-minute scans, while denoised images (DIs) were generated from 1-, 3-, 5-, and 10-minute scans. Image quality was assessed using normalized root mean square error (NRMSE), peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), visual analysis, and clinical metrics like BP_ND and ISR for diagnosis of non-neurodegenerative Parkinson disease (NPD), idiopathic PD (IPD), and atypical PD (APD). External validation used data from 2 hospitals with different scanners (hospital B: 1-, 3-, 5-, and 10-min; hospital C: 1-, 3-, and 5-min). In addition, motion artifact reduction was evaluated using the Dice similarity coefficient (DSC). Results: In hospital A, NRMSE, PSNR, and SSIM values improved with scan duration, with the 5-minute DIs achieving optimal quality (NRMSE 0.008, PSNR 42.13, SSIM 0.98). Visual analysis rated DIs from scans ≥3 minutes as adequate or higher. The mean BP_ND differences (95% CI) for each DIs were 0.19 (−0.01, 0.40), 0.11 (−0.02, 0.24), 0.08 (−0.03, 0.18), and 0.01 (−0.06, 0.07), with the CIs significantly decreasing. ISRs with the highest effect sizes for differentiating NPD, IPD, and APD (PP/AP, PP/VS, PC/VP) remained stable post-denoising. External validation showed 10-minute DIs (hospital B) and 1-minute DIs (hospital C) reached benchmarks of hospital A’s image quality metrics, with similar trends in visual analysis and BP_ND CIs. Furthermore, motion artifact correction in 9 patients yielded DSC improvements from 0.89 to 0.95 in striatal regions. Conclusions: The DL-model is capable of generating high-quality ¹⁸F FP-CIT PET images from shorter scans to enhance patient comfort, minimize motion artifacts, and maintain diagnostic precision. Furthermore, our study plays an important role in providing insights into how imaging quality assessment metrics can be used to determine the appropriate scan duration for different scanners with varying sensitivities.

Purpose: Convolutional neural networks (CNNs) have been studied for detecting bone metastases on bone scans; however, the application of ConvNeXt and transformer models has not yet been explored. This study aims to evaluate the performance of various deep learning models, including the ConvNeXt and transformer models, in diagnosing metastatic lesions from bone scans. Materials and Methods: We retrospectively analyzed bone scans from patients with cancer obtained at 2 institutions: the training and validation sets (n=4626) were from Hospital 1 and the test set (n=1428) was from Hospital 2. The deep learning models evaluated included ResNet18, the Data-Efficient Image Transformer (DeiT), the Vision Transformer (ViT Large 16), the Swin Transformer (Swin Base), and ConvNeXt Large. Gradient-weighted class activation mapping (Grad-CAM) was used for visualization. Results: Both the validation set and the test set demonstrated that the ConvNeXt large model (0.969 and 0.885, respectively) exhibited the best performance, followed by the Swin Base model (0.965 and 0.840, respectively), both of which significantly outperformed ResNet (0.892 and 0.725, respectively). Subgroup analyses revealed that all the models demonstrated greater diagnostic accuracy for patients with polymetastasis compared with those with oligometastasis. Grad-CAM visualization revealed that the ConvNeXt Large model focused more on identifying local lesions, whereas the Swin Base model focused on global areas such as the axial skeleton and pelvis. Conclusions: Compared with traditional CNN and transformer models, the ConvNeXt model demonstrated superior diagnostic performance in detecting bone metastases from bone scans, especially in cases of polymetastasis, suggesting its potential in medical image analysis.

Emerging evidence from observational studies suggests that lifestyle modifications, particularly moderate-intensity exercise, may confer neuroprotective benefits against dementia, potentially by enhancing brain resistance through clearance mechanisms. Using light-sheet fluorescence microscopy (LSFM) with tissue clearing, we investigated the role of voluntary swimming in ameliorating β-amyloid pathology in a transgenic Alzheimer’s disease (AD) mouse model. Twenty 52-week-old hAPPsw mice were randomly divided into a 5-week voluntary swimming intervention group and a control group (each n = 10). Each session included a 10-min swim followed by a 10-min rest, escalating from one session per day in the first week to three sessions per day by the fifth week. The excised brains were prepared using tissue-clearing and volume immunostaining with thioflavin-S for β-amyloid. For LSFM imaging, the individual plaque area and volume, total plaque load, and morphological parameters were quantified via an Imaris-based three-dimensional (3D) volumetric surface model. Visual comparison revealed that the intervention group presented significantly lower β-amyloid accumulation. The total surface volume of β-amyloid accumulation in the intervention group was significantly lower than that of the control group (intervention, 122,180,948 μm³ [105,854,660–169,063,081]; control, 167,201,016 μm³ [139,367,765–193,535,450]; p = 0.043). There were no significant differences in the morphological parameters, such as ellipticity and sphericity. Our LSFM study demonstrated notable reductions in β-amyloid, as evidenced by a decrease in total surface volume, in 52-week-old transgenic mice after a 5-week structured swimming program, supporting the notion that even in advanced AD stages, leisure-time voluntary swimming serves as an efficacious intervention for augmenting resistance to pathology.