STShumpei Takinami

Shumpei Takinami

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Shumpei Takinami has a total of 8 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    An Open-Source Horizontal Strabismus Simulator as an Evaluation Platform for Monocular Gaze Estimation Using Deep Learning Models

    by Shumpei Takinami, Yuka Morita, Jun Seita, Tetsuro Oshika

    Strabismus affects 2–4% of the global population, with horizontal cases accounting for more than 90%. Automated screening using monocular gaze estimation technology shows promise for early detection. However, existing models assume normal binocular vision, and their applicability to strabismus remains unvalidated due to the lack of evaluation platforms capable of reproducing disconjugate eye movements with known ground-truth angles. To address this gap, we developed an open-source, low-cost (approximately 200 USD) horizontal strabismus simulator. The simulator features two independently controllable artificial eyeballs mounted on a two-axis gimbal mechanism with servo motors and gyro sensors for real-time angle measurement. Mechanical accuracy achieved a mean absolute error of less than 0.1° across all axes, well below the clinical detection threshold of 1 prism diopter (≈0.57°). An evaluation of three representative AI models (Single Eye, GazeNet, and EyeNet) revealed estimation errors of 6.44–8.75°, substantially exceeding the clinical target of 2.8°. At this error level, small-angle strabismus (<15 prism diopters) would likely be missed, underscoring the need for strabismus-specific model development. Moreover, rapid accuracy degradation was observed beyond ±15° gaze angles. This platform establishes baseline performance metrics and provides a foundation for advancing gaze estimation technology for strabismus screening.


  • Open Access

    Pre-trained Vision Transformer With Masked Autoencoder for Automated Diabetic Macular Edema Detection from Optical Coherence Tomography Images

    by Shumpei Takinami, Shohei Morikawa, Tetsuro Oshika

    Purpose To develop and evaluate a novel self-supervised learning approach using Masked Autoencoder (MAE) pre-trained Vision Transformer (ViT) for automated detection of diabetic macular edema (DME) from optical coherence tomography (OCT) images, addressing the critical need for scalable screening solutions in diabetic eye care. Study Design Artificial intelligence model training. Methods We utilized the publicly available Kermany dataset containing 109,312 OCT images, defining DME detection as a binary classification task (11,559 DME vs. 97,753 non-DME images). Five deep learning architectures were compared: MAE-pretrained ViT (MAE_ViT), standard ViT, ResNet18, VGG19_bn, and EfficientNetV2. MAE_ViT underwent two-stage training: (1) self-supervised pre-training with 75% patch masking for 1,000 epochs to learn robust visual representations, and (2) supervised fine-tuning for DME classification. Model performance was evaluated using accuracy, sensitivity, specificity, F1 score, and area under the receiver operating characteristic curve (AU-ROC) with 95% confidence intervals calculated via bootstrap resampling. Results MAE_ViT achieved superior performance with AU-ROC 0.999 (95% CI: 0.999-1.000), accuracy 98.5% (95% CI: 97.7-99.2%), sensitivity 99.6% (95% CI: 98.7-100%), and specificity 98.1% (95% CI: 97.2-99.1%). VGG19_bn showed the second-best performance (AU-ROC 0.997), while ResNet18 demonstrated poor specificity (28.3%) despite perfect sensitivity. The self-supervised approach of MAE_ViT outperformed standard supervised ViT (AU-ROC 0.995), demonstrating the effectiveness of learning from unlabeled data. Conclusion MAE pre-trained Vision Transformer establishes a new benchmark for automated DME detection, offering exceptional diagnostic accuracy and potential for deployment in resource-constrained settings through reduced annotation requirements.