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.

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.

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: To report a case of paraneoplastic autoimmune retinopathy in a patient with a history of testicular cancer.Case summary: A 29-year-old man presented with photopsia and floaters. Initial fundus examination revealed no abnormal findings. However, he returned 1 month later with complaints of visual field defects. Fundus examination revealed diffuse white spots in the macula and midperipheral retina and fundus autofluorescence demonstrated hyper-autofluorescence. Optical coherence tomography showed disruption of the ellipsoid zone sparing the fovea. Visual field examination revealed peripheral visual field defects and an electroretinogram showed reduced rod and cone cell responses. Considering his history of testicular cancer, serum paraneoplastic autoantibody panel testing was performed which revealed borderline levels of anti-recoverin antibody leading to a diagnosis of paraneoplastic autoimmune retinopathy. The patient was treated with oral steroids and mycophenolate mofetil for 1 year. However, there was no improvement in the subjective symptoms or ophthalmologic findings. Conclusions This case of paraneoplastic autoimmune retinopathy in a young man with a history of testicular cancer highlights an early clinical presentation of the disease. It is crucial to recognize that the initial clinical presentation of autoimmune retinopathy can be nonspecific.

Purpose: To report a case of paraneoplastic autoimmune retinopathy in a patient with a history of testicular cancer.Case summary: A 29-year-old man presented with photopsia and floaters. Initial fundus examination revealed no abnormal findings. However, he returned 1 month later with complaints of visual field defects. Fundus examination revealed diffuse white spots in the macula and midperipheral retina and fundus autofluorescence demonstrated hyper-autofluorescence. Optical coherence tomography showed disruption of the ellipsoid zone sparing the fovea. Visual field examination revealed peripheral visual field defects and an electroretinogram showed reduced rod and cone cell responses. Considering his history of testicular cancer, serum paraneoplastic autoantibody panel testing was performed which revealed borderline levels of anti-recoverin antibody leading to a diagnosis of paraneoplastic autoimmune retinopathy. The patient was treated with oral steroids and mycophenolate mofetil for 1 year. However, there was no improvement in the subjective symptoms or ophthalmologic findings. Conclusions This case of paraneoplastic autoimmune retinopathy in a young man with a history of testicular cancer highlights an early clinical presentation of the disease. It is crucial to recognize that the initial clinical presentation of autoimmune retinopathy can be nonspecific.

Purpose: To report a case of paraneoplastic autoimmune retinopathy in a patient with a history of testicular cancer.Case summary: A 29-year-old man presented with photopsia and floaters. Initial fundus examination revealed no abnormal findings. However, he returned 1 month later with complaints of visual field defects. Fundus examination revealed diffuse white spots in the macula and midperipheral retina and fundus autofluorescence demonstrated hyper-autofluorescence. Optical coherence tomography showed disruption of the ellipsoid zone sparing the fovea. Visual field examination revealed peripheral visual field defects and an electroretinogram showed reduced rod and cone cell responses. Considering his history of testicular cancer, serum paraneoplastic autoantibody panel testing was performed which revealed borderline levels of anti-recoverin antibody leading to a diagnosis of paraneoplastic autoimmune retinopathy. The patient was treated with oral steroids and mycophenolate mofetil for 1 year. However, there was no improvement in the subjective symptoms or ophthalmologic findings. Conclusions This case of paraneoplastic autoimmune retinopathy in a young man with a history of testicular cancer highlights an early clinical presentation of the disease. It is crucial to recognize that the initial clinical presentation of autoimmune retinopathy can be nonspecific.

Purpose: To compare the effectiveness of warm saline and anti-fog solution for preventing fogging of a non-contact wide-field viewing system during vitrectomy. Methods: Five liters of water at 36°C were placed in a transparent container. The fogging areas of wide-field lenses were microscopically measured. We created three groups: lenses soaked in normal saline at 25°C for 1 minute (control), lenses soaked in normal saline at 50°C for 1 minute (warm saline), and lenses that were wiped with a sponge soaked in anti-fog solution (ULTRASTOP pro med. Solution, Sigmapharm, Vienna, Austria) after prior soaking in normal saline at 25°C for 1 minute (anti-fog). Images of fogged areas were acquired at 10 seconds and 1, 3, and 5 minutes. Extent of fogged areas and central lens invasion were determined. All experiments were repeated 10 times. Results: In the control group, the entire areas were always completely fogged. The average fog coverage values were 4.34 ± 1.28, 6.30 ± 1.38, 56.00 ± 25.01, and 93.81 ± 5.88% at 10 seconds and 1, 3, and 5 minutes in the warm saline group and 4.74 ± 0.57, 7.35 ± 0.96, 10.13 ± 1.09, and 11.74 ± 1.74% in the anti-fog group, respectively. There were significant differences at 3 and 5 minutes (p = 0.029, p = 0.012). Fogging of the central lens was detected in 8 tests after 3 minutes and all 10 tests after 5 minutes in the warm saline group, but no fogging was detected in the anti-fog group. Conclusions Application of an anti-fog solution to a wide-field viewing lens prevents lens fogging during vitrectomy.