Background/Objectives: Artificial intelligence (AI) has rapidly expanded across medical imaging with proposed applications in diagnosis, prognostication, and surgical planning. Concerns remain regarding methodological robustness and clinical readiness for many published models. This systematic review aimed to conduct a methodological audit of AI imaging studies relevant to contemporary neurosurgical practice—including intracranial, cerebrovascular, spinal, and connectomics-based applications—published in 2025. Methods: Following PRISMA guidelines and PROSPERO registration (CRD420261284068), PubMed was searched for studies published in 2025 evaluating machine learning or deep learning applications in MRI- or CT-based imaging. Three reviewers independently extracted data on validation strategy, data leakage risk, human comparator use, calibration reporting, and CLAIM/TRIPOD-AI adherence. Risk of bias was assessed using PROBAST+AI. Results: Of 1776 screened records, 91 studies met the inclusion criteria. China led contributions (54.9%), oncology was the most common domain (37.4%), and MRI was the predominant modality (67.0%). External validation was reported in 75.8% of studies, and 66.0% used multicenter cohorts. Data leakage risk was low in 93.4%. However, only 18.7% included human comparators, calibration was reported in 30.8%, and none achieved full CLAIM/TRIPOD-AI compliance. Conclusions: AI imaging studies published in 2025 demonstrate encouraging progress in multicenter design and external validation. However, persistent gaps in human benchmarking, calibration, and reporting suggest further methodological development is needed.

Background/Objectives: Intrahepatic cholestasis of pregnancy (ICP) is associated with adverse perinatal outcomes. However, its metabolic consequences on newborns remain inadequately characterized. This study investigated amino acid, carnitine, and acylcarnitine profiles in neonates born to mothers with ICP. Methods: This retrospective study encompassed 299 neonates born to mothers with ICP. For comparative analysis, term infants without additional complications (ICP-term, n = 150) were compared with term controls (n = 150). Capillary blood samples collected at 24–48 h of life as part of newborn screening were analyzed using LC–MS/MS for acylcarnitine and amino acid profiles. Results: The ICP cohort exhibited a high preterm delivery rate (46.2%), with maternal bile acids negatively correlating with gestational age (r = −0.266, p < 0.001). No inborn errors of metabolism were observed. Elevated levels of amino acids (alanine, leucine/isoleucine, valine, tyrosine, arginine, glycine, and ornithine) and specific acylcarnitines (C5, C5-OH, C10:1, and C18:2), along with decreased levels of amino acids (argininosuccinic acid and glutamic acid) and specific acylcarnitines (C3, C5-DC, C6-DC, C14, C14:1, C16, C16:1, and C18:1-OH), were observed in ICP-term neonates (p < 0.05). Receiver operating characteristic curve analysis identified ornithine (area under the curve [AUC] = 0.74) and leucine/isoleucine (AUC = 0.73) as strong discriminators. A multivariable model integrating multiple metabolites achieved high accuracy (AUC = 0.86 ± 0.03). Conclusions: This first comprehensive characterization of neonatal metabolic alterations in ICP reveals amino acid metabolism, fatty acid oxidation, and mitochondrial function disruptions, suggesting fetal adaptation to a cholestatic intrauterine environment. Metabolomic profiling may improve understanding of maternal–fetal interactions and inform strategies for risk stratification and long-term monitoring.

Background: Data extraction for systematic reviews is highly resource-intensive. This study evaluated four frontier large language models (LLMs) on complex structured metadata extraction from specialized neuroimaging artificial intelligence (AI) literature to determine their performance in automated evidence synthesis. Methods: We compared Google Gemini 3 Pro Preview, Anthropic Claude Opus 4.5, Perplexity Sonar Pro, and OpenAI GPT 5.2. Using a standardized prompt, each model extracted 22 variables from 91 peer-reviewed neuroimaging AI articles. The variables were stratified into low-, medium-, and high-complexity tiers. The performance was measured via the exact-match accuracy against a consensus-based expert ground truth. Results: The overall exact-match accuracy was moderate. Gemini 3 Pro Preview achieved the highest overall rate (56.4%), followed by Sonar Pro (52.1%), Claude Opus 4.5 (51.3%), and GPT 5.2 (46.5%). Gemini significantly outperformed all other models (p < 0.001). The performance declined dramatically as the variable complexity increased. Across models, the accuracy was 88.9–92.9% for low-complexity categorical fields, 47.0–63.3% for medium-complexity text extraction, and 2.7–15.5% for high-complexity variables requiring clinical judgment or multi-section synthesis. The most common type of error was misclassification. All four models scored 0% on the main performance metric, but this reflected a representational mismatch with the ground truth rather than extraction failure, indicating that the exact-match accuracy underestimates the true semantic performance. Conclusions: Frontier LLMs can effectively automate the retrieval of simple categorical data, but have serious difficulties with methodological variables that are complex. Although extraction can be fully automated for low-complexity fields, human review remains essential for context-dependent variables that require clinical judgment.

Objective: Caffeine is a proven medication used for the prevention and treatment of apnea in premature infants, offering both short- and long-term benefits. International guidelines provide a range of recommendations regarding the preterm population eligible for caffeine prophylaxis, including the timing, dosage, and duration of treatment. Our national guidelines, published prior to the most recent updates of the international guidelines, recommend the use of caffeine citrate starting from the first day after delivery for preterm infants with a gestational age of <28 weeks. For infants up to 32 weeks, if positive pressure ventilation is required, the decision should be made on an individual basis. This study aims to describe the variability in caffeine usage across neonatal intensive care units in our country.Methods: An online survey was sent to neonatologist who are members of the Turkish Neonatology Society to describe the variability in caffeine usage in neonatal intensive care units in our country.Results: We collected responses from 74 units. Prophylactic caffeine usage was observed as; GA ≤27^6/7: 98.6%, GA 28^0/7–28^6/7: 89.0%, GA 29^0/7–29^6/7: 75.3%, GA 30^0/7–31^6/7: 53.4%. 62.2% of units reported administering loading dose within the first two hours. The initial maintenance dose was 5 mg/kg in 64.8% of units, 10 mg/kg in 32.4% of units, and intermediate dose in 5.3% of units. 47.3% of units reported no routine dose adjustment. The postmenstrual age that caffeine treatment was stopped was found to be 34 (min-max; 32–36) weeks for infants without apnea and respiratory support, 36 (min-max; 34–52) weeks for infants without apnea but any respiratory support. The time to discharge after treatment cessation was found as; 1–4 days: 37.8%, 5–7 days: 68.9%. Among the 56 units with multiple responsible physicians, 32.1% reported intra-unit variations.Conclusion: The significant differences in caffeine usage characteristics between and within units highlight the need for clear recommendations provided by standardized guidelines.

Background: The reliable integration of large language models (LLMs) into neuroimaging data extraction workflows remains unresolved. Prior benchmarking shows that exact-match accuracy underestimates LLM extraction performance, but whether inter-model consensus and variable complexity can guide automation remains unclear. We evaluated whether inter-model consensus can serve as a confidence signal for human–artificial intelligence (AI) extraction and can guide complexity-stratified workflow triage. Methods: Four frontier LLMs were queried via OpenRouter with an identical zero-shot structured prompt to extract 22 predefined variables from 91 peer-reviewed neuroimaging AI articles, yielding 2002 article–variable items per model. Variables were stratified a priori into low- (n = 7), medium- (n = 8), and high-complexity (n = 7). Performance was compared with an expert reference using exact-match and semantic-equivalence accuracy. Item-level consensus and five triage strategies characterized the efficiency–accuracy trade-off. Results: Semantic-equivalence accuracy converged to 80.5–83.4% across models despite approximately ten percentage-point exact-match differences. Unanimous 4/4 consensus occurred in 45.6% (910/1994) of items, with exact-match accuracy of 85.8%, rising to 95.3% after semantic normalization; however, 14.2% still failed to match the reference. Reliability was complexity-dependent: 96.6% for low-complexity variables, 73.2% for medium-complexity variables, and 38.1% for high-complexity variables. A hybrid strategy auto-accepting 4/4 items and routing 3/4 items to rapid verification reduced estimated review effort by approximately 59%. Conclusions: Inter-model consensus is useful, but it is incomplete and depends on variable complexity. We show that LLM-assisted extraction in neuroimaging AI is a complexity-stratified workflow design problem: low-complexity neuroimaging variables may be selectively automated, while medium-complexity variables require rapid verification, and high-complexity methodological variables should remain human-led.