Background: Machine-learning (ML) and radiomics features have been utilized for survival outcome analysis in various cancers. This study aims to investigate the application of ML based on patients’ clinical features and radiomics features derived from bone scintigraphy (BS) and to evaluate recurrence-free survival in local or locally advanced prostate cancer (PCa) patients after the initial treatment. Methods: A total of 354 patients who met the eligibility criteria were analyzed and used to train the model. Clinical information and radiomics features of BS were obtained. Survival-related clinical features and radiomics features were included in the ML model training. Using the pyradiomics software, 128 radiomics features from each BS image’s region of interest, validated by experts, were extracted. Four textural matrices were also calculated: GLCM, NGLDM, GLRLM, and GLSZM. Five training models (Logistic Regression, Naive Bayes, Random Forest, Support Vector Classification, and XGBoost) were applied using K-fold cross-validation. Recurrence was defined as either a rise in PSA levels, radiographic progression, or death. To assess the classifier’s effectiveness, the ROC curve area and confusion matrix were employed. Results: Of the 354 patients, 101 patients were categorized into the recurrence group with more advanced disease status compared to the non-recurrence group. Key clinical features including tumor stage, radical prostatectomy, initial PSA, Gleason Score primary pattern, and radiotherapy were used for model training. Random Forest (RF) was the best-performing model, with a sensitivity of 0.81, specificity of 0.87, and accuracy of 0.85. The ROC curve analysis showed that predictions from RF outperformed predictions from other ML models with a final AUC of 0.94 and a p-value of <0.001. The other models had accuracy ranges from 0.52 to 0.78 and AUC ranges from 0.67 to 0.84. Conclusions: The study showed that ML based on clinical features and radiomics features of BS improves the prediction of PCa recurrence after initial treatment. These findings highlight the added value of ML techniques for risk classification in PCa based on clinical features and radiomics features of BS.

Background: ChatGPT has the potential to enhance patient education by offering clear and accurate responses, but its reliability in providing precise medical information is still under investigation. This study evaluates the effectiveness in assisting healthcare professionals with patient inquiries about radioiodine therapy. Methods: This study used OpenAI's GPT-4o and GPT-4 models, with each query submitted as a separate prompt. Chain-of-thought prompting was utilized to require the model to articulate its step-by-step reasoning prior to the final answer, thereby making the decision process transparent for qualitative evaluation. Three responses were generated per prompt and evaluated by three nuclear medicine doctors using a 4-point Likert scale across five aspects: Appropriateness, Helpfulness, Consistency, Validity of References, and Empathy. Normality test, Wilcoxon signed-rank test, and chi-square tests were used for analysis. Results: A total of 126 paired responses from GPT-4 and GPT-4o were independently rated by three nuclear-medicine physicians. Both models performed similarly across the main dimensions—appropriateness, helpfulness, consistency, and validity of reference—with no statistically significant differences (Wilcoxon signed-rank, p ≥ 0.01). High-level ratings (score ≥ 3) were achieved in appropriateness for 90.4% of GPT-4 outputs and 84.9% of GPT-4o outputs, and in helpfulness for 92.1% of outputs from both models. Citation accuracy was limited: fully valid references were present in 20.6% of GPT-4 and 21.4% of GPT-4o responses. Empathy was judged present in 56.3% of GPT-4 and 66.7% of GPT-4o answers (χ², p > 0.05). There was low inter-rater agreement (Fleiss κ = 0.04). Conclusion: The results suggest that ChatGPT can furnish generally appropriate and helpful answers to frequently asked questions in radioactive iodine treatment, yet citation accuracy remains limited, underscoring the need for clinician oversight. GPT-4o and GPT-4 demonstrated comparable performance, indicating that model selection within this family has minimal impact under the controlled conditions studied.

This study aimed to develop and evaluate an AI-driven platform, the Adaptive RAG Assistant MRI Platform (ARAMP), for assisting in the diagnosis and reporting of brain metastases using post-contrast axial T1-weighted (AX_T1+C) MRI. In this retrospective study, 2447 cancer patients who underwent MRI between 2010 and 2022 were screened. A subset of 100 randomized patients with confirmed brain metastases and 100 matched non-cancer controls were selected for evaluation. ARAMP integrates quantitative radiomic feature extraction with an adaptive Retrieval-Augmented Generation (RAG) framework based on a large language model (LLM, GPT-4o), incorporating five authoritative medical references. Three board-certified neuroradiologists and an independent LLM (Gemini 2.0 Pro) assessed ARAMP performance. Metrics of the assessment included Pre-/Post-Trained Inference Difference, Inter-Inference Agreement, and Sensitivity. Post-training, ARAMP achieved a mean Inference Similarity score of 67.45%. Inter-Inference Agreement among radiologists averaged 30.20% (p = 0.01). Sensitivity for brain metastasis detection improved from 0.84 (pre-training) to 0.98 (post-training). ARAMP also showed improved reliability in identifying brain metastases as the primary diagnosis post-RAG integration. This adaptive RAG-based framework may improve diagnostic efficiency and standardization in radiological workflows.

Background and Purpose: Neurology requires integration of clinical reasoning, imaging interpretation, and current knowledge, making it an ideal field for evaluating large language models (LLMs). Methods: Using 1715 questions from the Taiwan Neurology Board Examination (2018–2024), we assessed four LLMs—GPT-4o, GPT-o1, DeepSeek-V3, and DeepSeek-R1—across four formats: single-choice, multiple-choice, true–false, and image-based items. Results: GPT-o1 achieved the highest overall accuracy (83.86%) and demonstrated strong performance on cognitively demanding tasks (82.50% on true–false; 77.26% on image-based). DeepSeek-V3 scored lowest (65.62%) and showed the greatest variability. Statistical analyses confirmed significant inter-model differences (p < 0.01). Accuracy declined across all models in 2024, coinciding with shifts in question design. DeepSeek-R1 was further penalized by alignment-based refusals, resulting in up to 3.81% score loss. Conclusions: These results position the Taiwan Neurology Board Exam as a rigorous benchmark for LLM evaluation and underscore GPT-o1’s potential utility in neurology education and decision support.