Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Purpose: This study aimed to develop a magnetic resonance imaging (MRI)–based radiomics model to predict high-risk pathologic features for lung adenocarcinoma: micropapillary and solid pattern (MPsol), spread through air space, and poorly differentiated patterns. Materials and Methods: As a prospective study, we screened clinical N0 lung cancer patients who were surgical candidates and had undergone both 18F-fluorodeoxyglucose (FDG) positron emission tomography–computed tomography (PET/CT) and chest CT from August 2018 to January 2020. We recruited patients meeting our proposed imaging criteria indicating high-risk, that is, poorer prognosis of lung adenocarcinoma, using CT and FDG PET/CT. If possible, these patients underwent an MRI examination from which we extracted 77 radiomics features from T1-contrast-enhanced and T2-weighted images. Additionally, patient demographics, maximum standardized uptake value on FDG PET/CT, and the mean apparent diffusion coefficient value on diffusion-weighted image, were considered together to build prediction models for high-risk pathologic features. Results: Among 616 patients, 72 patients met the imaging criteria for high-risk lung cancer and underwent lung MRI. The magnetic resonance (MR)–eligible group showed a higher prevalence of nodal upstaging (29.2% vs. 4.2%, p < 0.001), vascular invasion (6.5% vs. 2.1%, p=0.011), high-grade pathologic features (p < 0.001), worse 4-year disease-free survival (p < 0.001) compared with non-MR-eligible group. The prediction power for MR-based radiomics model predicting high-risk pathologic features was good, with mean area under the receiver operating curve (AUC) value measuring 0.751-0.886 in test sets. Adding clinical variables increased the predictive performance for MPsol and the poorly differentiated pattern using the 2021 grading system (AUC, 0.860 and 0.907, respectively). Conclusion Our imaging criteria can effectively screen high-risk lung cancer patients and predict high-risk pathologic features by our MR-based prediction model using radiomics.

Purpose: This study aimed to develop a magnetic resonance imaging (MRI)–based radiomics model to predict high-risk pathologic features for lung adenocarcinoma: micropapillary and solid pattern (MPsol), spread through air space, and poorly differentiated patterns. Materials and Methods: As a prospective study, we screened clinical N0 lung cancer patients who were surgical candidates and had undergone both 18F-fluorodeoxyglucose (FDG) positron emission tomography–computed tomography (PET/CT) and chest CT from August 2018 to January 2020. We recruited patients meeting our proposed imaging criteria indicating high-risk, that is, poorer prognosis of lung adenocarcinoma, using CT and FDG PET/CT. If possible, these patients underwent an MRI examination from which we extracted 77 radiomics features from T1-contrast-enhanced and T2-weighted images. Additionally, patient demographics, maximum standardized uptake value on FDG PET/CT, and the mean apparent diffusion coefficient value on diffusion-weighted image, were considered together to build prediction models for high-risk pathologic features. Results: Among 616 patients, 72 patients met the imaging criteria for high-risk lung cancer and underwent lung MRI. The magnetic resonance (MR)–eligible group showed a higher prevalence of nodal upstaging (29.2% vs. 4.2%, p < 0.001), vascular invasion (6.5% vs. 2.1%, p=0.011), high-grade pathologic features (p < 0.001), worse 4-year disease-free survival (p < 0.001) compared with non-MR-eligible group. The prediction power for MR-based radiomics model predicting high-risk pathologic features was good, with mean area under the receiver operating curve (AUC) value measuring 0.751-0.886 in test sets. Adding clinical variables increased the predictive performance for MPsol and the poorly differentiated pattern using the 2021 grading system (AUC, 0.860 and 0.907, respectively). Conclusion Our imaging criteria can effectively screen high-risk lung cancer patients and predict high-risk pathologic features by our MR-based prediction model using radiomics.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Purpose: This study aimed to develop a magnetic resonance imaging (MRI)–based radiomics model to predict high-risk pathologic features for lung adenocarcinoma: micropapillary and solid pattern (MPsol), spread through air space, and poorly differentiated patterns. Materials and Methods: As a prospective study, we screened clinical N0 lung cancer patients who were surgical candidates and had undergone both 18F-fluorodeoxyglucose (FDG) positron emission tomography–computed tomography (PET/CT) and chest CT from August 2018 to January 2020. We recruited patients meeting our proposed imaging criteria indicating high-risk, that is, poorer prognosis of lung adenocarcinoma, using CT and FDG PET/CT. If possible, these patients underwent an MRI examination from which we extracted 77 radiomics features from T1-contrast-enhanced and T2-weighted images. Additionally, patient demographics, maximum standardized uptake value on FDG PET/CT, and the mean apparent diffusion coefficient value on diffusion-weighted image, were considered together to build prediction models for high-risk pathologic features. Results: Among 616 patients, 72 patients met the imaging criteria for high-risk lung cancer and underwent lung MRI. The magnetic resonance (MR)–eligible group showed a higher prevalence of nodal upstaging (29.2% vs. 4.2%, p < 0.001), vascular invasion (6.5% vs. 2.1%, p=0.011), high-grade pathologic features (p < 0.001), worse 4-year disease-free survival (p < 0.001) compared with non-MR-eligible group. The prediction power for MR-based radiomics model predicting high-risk pathologic features was good, with mean area under the receiver operating curve (AUC) value measuring 0.751-0.886 in test sets. Adding clinical variables increased the predictive performance for MPsol and the poorly differentiated pattern using the 2021 grading system (AUC, 0.860 and 0.907, respectively). Conclusion Our imaging criteria can effectively screen high-risk lung cancer patients and predict high-risk pathologic features by our MR-based prediction model using radiomics.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

The healing potential of the meniscus is limited because of the complex morphological features and insufficient blood supply. Various surgical procedures are used to treat meniscal lesions, such as partial meniscectomy, repair, and allograft transplantation. On the other hand, such strategies cannot fully regenerate and restore the morphology and functional aspects of the meniscus. This paper reviews the key aspects concerning meniscus regeneration, biological augmentation strategies, meniscus replacement scaffolds, cell therapies, and emerging tissue engineering technologies.