Announcements

As Japan’s population increasingly ages, the number of bedridden patients who need long-term care is also rising. However, it is difficult for these patients to travel to distant places. Therefore, in order to enhance their quality of life (QOL), we developed a remote-controlled drone system that only uses the movement of their eyes. This paper presents clinical experimental results that verify the effectiveness of this drone system for bedridden patients. The participants of this study included bedridden patients and patients with relatively high levels of independence in their activities of daily living (ADL). The experimental results show the efficacy of operating the drone in a distant place with only the eyes, seeing the scenery of the distant place with a camera installed on the drone, and communicating with local people.

Both the human eye and a digital camera sense both light intensity and color. We use visual information for various purposes, including seeing things, keeping our physical balance, and making judgments based on one's situational experiences. It naturally becomes subjective and uncertain. The quantification of color from digital images, including prints and static/dynamic digital data, is necessary to reduce uncertainty and subjectivity for quantified data-driven judgments and control applications in various fields. This study shows reliable, image-based quantification of color using specially developed image processing and analysis software. This image-based quantification software (PicMan) is useful for a wide range of applications in art, fashion, medical science, and many other fields.

Accurate prediction of electrical loads is crucial for efficient power system operation and market management. Various forecasting platforms have been proposed to address this challenge, including the use of recurrent neural networks trained on hourly or daily load inputs. This paper presents a framework that employs an RNN model to forecast future electrical loads and provides comparative analysis with other state-of-the-art architectures trained in different variations. Extensive testing on a dataset including Greece's electricity load values per hour demonstrates the framework's ability to capture underlying patterns and achieve high predictive accuracy. Notably, the proposed RNN outperforms more complex neural networks, indicating its effectiveness in capturing data patterns or trends.

The integration of satellite technology with cellular mobile networks has evolved as an essential pursuit in achieving global wireless access. The rise of LTE and 5G users prioritizing seamless coverage, especially in remote areas, has propelled the development of direct-to-cellular 5G satellite NTN. This study explores the challenges and feasibility of integrating 5G–6G satellite networks within existing terrestrial and satellite frequency bands. Two approaches are examined: spectrum convergence within terrestrial bands and the utilization of dedicated NTN frequencies. Through detailed interference analysis, this study sheds light on the compatibility hurdles that stakeholders must address to harness the potential of n255 and n256 bands for future 5G-6G satellite technologies.

A new deep neural model called MS-CNN has been developed, and this model has achieved state-of-the-art accuracy (96.05%) in classifying six lung diseases from chest X-ray images. MS-CNN is evaluated on a dataset of 6650 chest X-ray images, outperforms previously reported models and could help in improving early diagnosis and treatment for patients with various lung diseases. The research also showed that MS-CNN can explain its predictions using techniques such as SHAP and Grad-CAM, which could help clinicians in better understanding how the model is making its decisions. The findings of this study suggest that MS-CNN has the potential to revolutionize the diagnosis of lung diseases. The model could be made available to clinicians in the near future, and it could have a significant impact on the lives of patients with lung diseases.

A deep learning classifier was developed for grading the degree of brain injury in hypoxic-ischemic encephalopathy (HIE) for full-term neonates. This utilized a fully convolutional architecture, trained to classify the brain injury grade based on a window of multichannel EEG. There was minimal pre-processing, and no need for the design of hand-crafted features; this demonstrated the ability of the deep architecture to extract features from the raw EEG time series. This classifier was tested in mismatched conditions on a large clinical dataset and provides state-of-the-art performance. Real-time grading of brain injury will facilitate clinical decision making in the treatment of HIE, including decisions regarding the initiation of cooling therapy.