Background/Objectives: To develop a DL-based model predicting recurrence risk in HER2-low breast cancer patients and to compare performance of the MRI-alone, clinicopathologic-alone, and combined models. Methods: We analyzed 453 patients with HER2-low breast cancer who underwent surgery and preoperative breast MRI between May 2018 and April 2022. Patients were randomly assigned to either a training cohort (n = 331) or a test cohort (n = 122). Imaging features were extracted from DCE-MRI and ADC maps, with regions of interest manually annotated by radiologists. Clinicopathological features included tumor size, nodal status, histological grade, and hormone receptor status. Three DL prediction models were developed: a CNN-based MRI-alone model, a clinicopathologic-alone model based on a multi-layer perceptron (MLP) and a combined model integrating CNN-extracted MRI features with clinicopathological data via MLP. Model performance was evaluated using AUC, sensitivity, specificity, and F1-score. Results: The MRI-alone model achieved an AUC of 0.69 (95% CI, 0.68–0.69), with a sensitivity of 37.6% (95% CI, 35.7–39.4), specificity of 87.5% (95% CI, 86.9–88.2), and F1-score of 0.34 (95% CI, 0.33–0.35). The clinicopathologic-alone model yielded the highest AUC of 0.92 (95% CI, 0.92–0.92) and sensitivity of 93.6% (95% CI, 93.4–93.8), but showed the lowest specificity (72.3%, 95% CI, 71.8–72.8) and F1-score of 0.50 (95% CI, 0.49–0.50). The combined model demonstrated the most balanced performance, achieving an AUC of 0.90 (95% CI, 0.89–0.91), sensitivity of 80.0% (95% CI, 78.7–81.3), specificity of 83.2% (95% CI: 82.7–83.6), and the highest F1-score of 0.55 (95% CI, 0.54–0.57). Conclusions: The DL-based model combining MRI and clinicopathological features showed superior performance in predicting recurrence in HER2-low breast cancer. This multimodal approach offers a framework for individualized risk assessment and may aid in refining follow-up strategies.

Transfer learning has emerged as a powerful technique in machine learning, wherein knowledge acquired from one task or dataset is leveraged to improve performance on related tasks and datasets. Although univariate time series forecasting plays a pivotal role in numerous real-world applications, the adoption of transfer learning in this domain has been relatively underexplored. This paper proposes a new transfer learning approach for univariate time series forecasting, combining the Neural Basis Expansion Analysis for interpretable Time Series forecasting (N-BEATS) and Multi-Layer Perceptron (MLP). Furthermore, the Boltzmann-weighted ensemble is proposed for prediction, serving as an advanced inference method that improves forecasting accuracy without incurring significant computational overhead. We constructed a pre-trained model for knowledge transfer. Then, we fine-tuned it to predict time series data from seven domains (temperature, solar radiation, electricity demand, air pollution, wind speed, traffic volume, and coin price). Our experimental results show that transfer learning enhances predictive performance, and investigations into additional layers reveal that MLP architectures are superior to long short-term memory and original transformer architectures in terms of Root Mean Square Error (RMSE), whereas the advanced time-series transformer PatchTST shows potential in transfer learning. Additionally, we conducted various experiments for qualitative analyses. The results of experiments consistently demonstrate the superior performance of transfer learning for univariate time series forecasting, offering valuable insights into designing more robust deep learning solutions in data-driven environments.

Objective: To determine whether repetition‐based saccadic fatigability measured with three‐dimensional video‐oculography (VOG) distinguishes myasthenia gravis (MG) from unilateral ocular motor cranial nerve palsy (CNP) in patients with diplopia. Methods: In this prospective cross‐sectional diagnostic study conducted from August 2022 to July 2024, consecutive patients with MG or unilateral third, fourth, or sixth CNP, and healthy controls underwent repetitive horizontal and vertical saccade testing (±°15 targets; 75 cycles at 0.25 Hz). Prespecified index metrics were percentage change in saccadic range and time to target acquisition from the second saccade to the mean of the last five saccades. Diagnostic discrimination was assessed using receiver operating characteristic (ROC) curves. Results: The cohort comprised 231 participants: 93 with MG, 54 with CNP and 84 healthy controls. MG showed greater repetition‐related range decrement and longer time to target acquisition than healthy controls and axis‐matched CNP. In the horizontal comparison (MG vs. sixth CNP), AUCs were 0.800 for percentage range change and 0.787 for percentage time‐to‐target change. In the vertical comparison (MG vs. third/fourth CNP), percentage range change showed the best discrimination (AUC 0.868, 95% CI 0.783 to 0.934); a threshold of −5.95% yielded 79.6% sensitivity and 87.5% specificity. CNP showed early baseline slowing, whereas MG showed progressive hypometria with relatively preserved peak velocity. Conclusions: Repetition‐based VOG captures a dynamic fatigability signature in MG that differs from the more static deficit of unilateral ocular motor CNP. Quantitative saccadic fatigability may provide a useful objective adjunct to the neuro‐ophthalmic evaluation of diplopia.