Development of AI-generated medical responses using the ChatGPT for cancer patients.

BACKGROUND AND OBJECTIVE: To develop a healthcare chatbot service (AI-guided bot) that conducts real-time conversations using large language models to provide accurate health information to patients. METHODS: To provide accurate and specialized medical responses, we integrated several cancer practice guidelines. The size of the integrated meta-dataset was 1.17 million tokens. The integrated and classified metadata were extracted, transformed into text, segmented to specific character lengths, and vectorized using the embedding model. The AI-guide bot was implemented using Python 3.9. To enhance the scalability and incorporate the integrated dataset, we combined the AI-guide bot with OpenAI and the LangChain framework. To generate user-friendly conversations, a language model was developed based on Chat-Generative Pretrained Transformer (ChatGPT), an interactive conversational chatbot powered by GPT-3.5. The AI-guide bot was implemented using ChatGPT3.5 from Sep. 2023 to Jan. 2024. RESULTS: The AI-guide bot allowed users to select their desired cancer type and language for conversational interactions. The AI-guided bot was designed to expand its capabilities to encompass multiple major cancer types. The performance of the AI-guide bot responses was 90.98 ± 4.02 (obtained by summing up the Likert scores). CONCLUSIONS: The AI-guide bot can provide medical information quickly and accurately to patients with cancer who are concerned about their health.

Characterization of 3D Organotypic Culture of Mouse Adipose-Derived Stem Cells.

Although stem cells are a promising avenue for harnessing the potential of adipose tissue, conventional two-dimensional (2D) culture methods have limitations. This study explored the use of three-dimensional (3D) cultures to preserve the regenerative potential of adipose-derived stem cells (ADSCs) and investigated their cellular properties. Flow cytometric analysis revealed significant variations in surface marker expressions between the two culture conditions. While 2D cultures showed robust surface marker expressions, 3D cultures exhibited reduced levels of CD44, CD90.2, and CD105. Adipogenic differentiation in 3D organotypic ADSCs faced challenges, with decreased organoid size and limited activation of adipogenesis-related genes. Key adipocyte markers, such as lipoprotein lipase (LPL) and adipoQ, were undetectable in 3D-cultured ADSCs, unlike positive controls in 2D-cultured mesenchymal stem cells (MSCs). Surprisingly, 3D-cultured ADSCs underwent mesenchymal-epithelial transition (MET), evidenced by increased E-cadherin and EpCAM expression and decreased mesenchymal markers. This study highlights successful ADSC organoid formation, notable MSC phenotype changes in 3D culture, adipogenic differentiation challenges, and a distinctive shift toward an epithelial-like state. These findings offer insights into the potential applications of 3D-cultured ADSCs in regenerative medicine, emphasizing the need for further exploration of underlying molecular mechanisms.

COL6A1 expression as a potential prognostic biomarker for risk stratification of T1 high grade bladder cancer: Unveiling the aggressive nature of a distinct non-muscle invasive subtype.

PURPOSE: T1 high grade (T1HG) bladder cancer (BC) is a type of non-muscle invasive BC (NMIBC) that is recognized as an aggressive subtype with a heightened propensity for progression. Current risk stratification methods for NMIBC rely on clinicopathological indicators; however, these approaches do not adequately capture the aggressive nature of T1HG BC. Thus, new, more accurate biomarkers for T1HG risk stratification are needed. Here, we enrolled three different patient cohorts and investigated expression of collagen type VI alpha 1 (COL6A1), a key component of the extracellular matrix, at different stages and grades of BC, with a specific focus on T1HG BC. MATERIALS AND METHODS: Samples from 298 BC patients were subjected to RNA sequencing and real-time polymerase chain reaction. RESULTS: We found that T1HG BC and muscle invasive BC (MIBC) exhibited comparable expression of COL6A1, which was significantly higher than that by other NMIBC subtypes. In particular, T1HG patients who later progressed to MIBC had considerably higher expression of COL6A1 than Ta, T1 low grade patients, and patients that did not progress, highlighting the aggressive nature and higher risk of progression associated with T1HG BC. Moreover, Cox and Kaplan-Meier survival analyses revealed a significant association between elevated expression of COL6A1 and poor progression-free survival of T1HG BC patients (multivariate Cox hazard ratio, 16.812; 95% confidence interval, 3.283-86.095; p=0.001 and p=0.0002 [log-rank test]). CONCLUSIONS: These findings suggest that COL6A1 may be a promising biomarker for risk stratification of T1HG BC, offering valuable insight into disease prognosis and guidance of personalized treatment decisions.

Multitask Learning-Based Deep Signal Identification for Advanced Spectrum Sensing.

The explosive demand for wireless communications has intensified the complexity of spectrum dynamics, particularly within unlicensed bands. To promote efficient spectrum utilization and minimize interference during communication, spectrum sensing needs to evolve to a stage capable of detecting multidimensional spectrum states. Signal identification, which identifies each device's signal source, is a potent method for deriving the spectrum usage characteristics of wireless devices. However, most existing signal identification methods mainly focus on signal classification or modulation classification, thus offering limited spectrum information. In this paper, we propose DSINet, a multitask learning-based deep signal identification network for advanced spectrum sensing systems. DSINet addresses the deep signal identification problem, which involves not only classifying signals but also deriving the spectrum usage characteristics of signals across various spectrum dimensions, including time, frequency, power, and code. Comparative analyses reveal that DSINet outperforms existing shallow signal identification models, with performance improvements of 3.3% for signal classification, 3.3% for hall detection, and 5.7% for modulation classification. In addition, DSINet solves four different tasks with a 65.5% smaller model size and 230% improved computational performance compared to single-task learning model sets, providing meaningful results in terms of practical use.