Publications

BACKGROUND: Dementia affects over 55 million people worldwide. Mild cognitive impairment (MCI) often precedes Alzheimer's disease (AD). Clinical management requires integrating uncertain evidence from neuropsychological testing, neuroimaging, and biomarkers. Large language models (LLMs) also generate probabilistic outputs, but whether they can reliably support diagnostic, therapeutic, or educational tasks in AD and MCI has not been systematically examined.

METHODS: We searched PubMed, Scopus, and PubMed Central (January 2023 to April 2026) for studies evaluating generative LLMs on clinical tasks in Alzheimer's disease (AD) or mild cognitive impairment (MCI). Risk of bias was assessed using QUADAS-AI and AXIS. Narrative synthesis followed the SWiM guideline.

PROSPERO: CRD420261372436.

RESULTS: Eleven studies were included: diagnosis (n = 3), treatment guidance (n = 2), and patient/caregiver education (n = 8); two studies contributed to multiple domains. Diagnostic models achieved high internal accuracy (0.94-0.97) but declined on external validation; three-way classification accuracy dropped approximately 7% points, and MMSE-prediction R² collapsed from 0.90 to 0.25 on an external dataset. Treatment guidance approached but did not match structured clinical guidelines. Educational outputs were rated moderate to high quality but lacked source attribution and exceeded recommended reading levels; retrieval augmentation improved usability without improving accuracy. Hallucination was quantified in only 2 of 11 studies, and no study evaluated prospective clinical use.

CONCLUSIONS: Current evidence does not support the use of LLMs for diagnosis, treatment selection, or patient education in AD/MCI without clinician oversight. These findings reflect the specific model versions, prompting strategies, and evaluation conditions in place at the time of each study, and are further limited by small heterogeneous evaluations, sparse hallucination measurement, and absence of prospective clinical validation.

PURPOSE: To evaluate the impact of socioeconomic factors on survival in patients with HCC undergoing interventional LRT at a tertiary center.

MATERIALS AND METHODS: A retrospective cohort of 1,091 HCC patients treated with interventional LRT (2000-2022). Clinical and socioeconomic data (gender, insurance, race, education, language, marital status) were collected. Survival analysis utilized Kaplan-Meier and multivariable Cox regression to assess the impact of socioeconomic factors . Among identified patients (median age: 63 [IQR=12.14], 81% male), the majority had Medicare (43%) or private insurance (33%), were White (63%), had a college (42%) or high school (45%) education, and were married (51%). Median AFP was 11 (IQR=76); most were Child-Pugh B (72%), BCLC Stage A (66%) and ECOG 0 (75%). Overall mortality was 62%, and 23% underwent liver transplant.

RESULTS: Female gender (HR 1.26, 95% CI 1.03-1.55, p = 0.028), lower education (high school vs. college; HR 1.24, 95% CI 1.04-1.50, p = 0.015), and being divorced (HR 1.43, 95% CI 1.08-1.90, p = 0.012) were independently associated with increased mortality. Asian race predicted improved survival (HR 0.44, 95% CI 0.29-0.67, p < 0.05).

CONCLUSION: Asian race, male gender, marriage, and higher education independently predicted better survival in HCC patients receiving interventional LRT.

OBJECTIVE: To compare computed tomography (CT) scan-based morphometric measures associated with atlanto-occipital joint (AOJ) instability between Chiari malformation type I (CMI) patients and controls and to examine associations between these measures, clinical symptoms, and established craniocervical/posterior fossa morphometrics in CMI.

MATERIALS AND METHODS: Four AOJ-related morphometric measures were evaluated on CT: condyle-C1 interval (CCI), C1 socket depth-to-length ratio (DL ratio), condyle depth-to-depth ratio (CD ratio), and atlas tilt angle. Measurements were compared between CMI (n = 45) and controls (n = 55). Within CMI, associations were tested between AOJ measures and symptom variables, as well as 13 established craniocervical/posterior fossa morphometrics.

RESULTS: Mean CCI was slightly larger in CMI than in controls (1.00 vs. 0.91 mm, P < 0.05). DL ratio, CD ratio, and atlas tilt angle did not differ significantly between groups. Within CMI, CCI showed a modest association with hypermobility (r = 0.34, P < 0.05). Moderate correlations were observed between AOJ measures and selected morphometrics, suggesting relationships between AOJ geometry and posterior fossa/craniocervical anatomy.

CONCLUSION: Among the AOJ measures examined, only CCI demonstrated a small but statistically significant group difference and a modest association with hypermobility in CMI. The shape of the AOJ curvature was not different between CMI and control. However, further investigation of AOJ-related morphometrics as potential contributors to symptom heterogeneity among CMI patients is necessary.

PURPOSE: To evaluate the impact of reducing protective equipment requirements on post-procedural infection rates in ultrasound-guided paracenteses, comparing outcomes between full protective equipment protocol (sterile gloves, gown, surgical mask, and hat) vs. limited protective equipment protocol (sterile gloves only).

MATERIALS AND METHODS: A single-center retrospective study analyzed 1,177 consecutive ultrasound-guided paracenteses performed between June 1st, 2021 and May 31st, 2022. The study compared infection rates across two six-month periods: before (619 procedures) and after (558 procedures) the 12/1/2021 policy change reducing protective equipment requirements. Primary outcome measures included post-procedural infection rates, specifically spontaneous bacterial peritonitis (SBP), soft tissue infection, or other intra-abdominal infections, evaluated at 7 and 14 days post-procedure. Analysis was conducted at both the individual paracentesis and patient level.

RESULTS: No statistically significant differences were observed in post-procedural infection rates between the full and limited protective equipment groups. The 7-day infection rates were 1.1% vs. 0.9% (p=0.41), and 14-day infection rates were 1.5% vs. 1.8% (p=0.65) for full versus limited protective equipment groups, respectively. Patient-level analysis showed similar results, with 14-day infection rates of 3.3% vs. 4.1% (p=0.69). The median volume of fluid removed remained consistent between groups (3.2L vs. 3.0L, p=0.50).

CONCLUSION: Sterile gloves alone provide adequate protection for ultrasound-guided paracentesis while reducing the use of protective equipment, enabling simplified protocols without compromising patient safety.

OBJECTIVE: In patients with acute ischemic stroke, CT perfusion (CTP)-derived infarct core is valuable for prognostication, triage and transfer decision-making, and for informing studies of emerging therapeutic targets. In this study, we compare the accuracy and variability of the infarct core predicted by 2 FDA-cleared CTP programs, using diffusion-weighted MRI (DWI) as the reference standard.

METHODS: We analyzed 61 stroke patients who underwent admission CTP and DWI within 90 minutes. Infarct core was estimated using relative cerebral blood flow thresholds and compared with DWI-derived ground truth. After coregistration of CTP and DWI, Dice similarity coefficients were calculated to quantify the topographic concordance of the infarct core.

RESULTS: The CTP-determined infarct core volumes were significantly correlated with DWI but demonstrated substantial variability and bias. Our results showed limited topographic overlap, with median dice scores of 0.367 (CTP-A) and 0.289 (CTP-B). However, in patients with larger infarcts (volumes ≥50 mL), CTP provides more reliable estimates of spatial agreement, reaching median Dice scores of 0.61 (CTP-A) and 0.47 (CTP-B).

CONCLUSIONS: Although these findings show a generally low accuracy of CTP estimation, they suggest that CTP may offer clinically meaningful insights for decision-making in the large-core setting and inform the design of future trials.