Using direct radiologic image inputs, two large language models, GPT-4V and Gemini Pro Vision, improved in diagnostic accuracy with increasing temperature settings; although GPT-4V slightly underperformed compared with radiologists, it nonetheless demonstrated promising potential as a supportive tool in diagnostic decision-making.

Large language models accurately answered New England Journal of Medicine Image Challenge cases using radiologic inputs, outperforming a medical student; however, their accuracy decreased with shorter text lengths, regardless of image inputs.

BACKGROUND AND PURPOSE: Limited evidence has suggested that a deep learning automatic brain segmentation and classification method, based on T1-weighted brain MR images, can predict Alzheimer disease. Our aim was to develop and validate a deep learning–based automatic brain segmentation and classification algorithm for the diagnosis of Alzheimer disease using 3D T1-weighted brain MR images. MATERIALS AND METHODS: A deep learning–based algorithm was developed using a dataset of T1-weighted brain MR images in consecutive patients with Alzheimer disease and mild cognitive impairment. We developed a 2-step algorithm using a convolutional neural network to perform brain parcellation followed by 3 classifier techniques including XGBoost for disease prediction. All classification experiments were performed using 5-fold cross-validation. The diagnostic performance of the XGBoost method was compared with logistic regression and a linear Support Vector Machine by calculating their areas under the curve for differentiating Alzheimer disease from mild cognitive impairment and mild cognitive impairment from healthy controls. RESULTS: In a total of 4 datasets, 1099, 212, 711, and 705 eligible patients were included. Compared with the linear Support Vector Machine and logistic regression, XGBoost significantly improved the prediction of Alzheimer disease (P < .001). In terms of differentiating Alzheimer disease from mild cognitive impairment, the 3 algorithms resulted in areas under the curve of 0.758–0.825. XGBoost had a sensitivity of 68% and a specificity of 70%. In terms of differentiating mild cognitive impairment from the healthy control group, the 3 algorithms resulted in areas under the curve of 0.668–0.870. XGBoost had a sensitivity of 79% and a specificity of 80%. CONCLUSIONS: The deep learning–based automatic brain segmentation and classification algorithm allowed an accurate diagnosis of Alzheimer disease using T1-weighted brain MR images. The widespread availability of T1-weighted brain MR imaging suggests that this algorithm is a promising and widely applicable method for predicting Alzheimer disease.

Objectives To assess the incidence of ARIA in clinical trials of anti-Aβ immunotherapy and compare the incidence among different agents and clinical characteristics to identify possible predisposing factors for ARIA. Methods The PubMed and Embase databases were searched for clinical trials of anti-Aβ immunotherapy published on or before January 12, 2022. Phase 2 or 3 randomized controlled trials reporting detailed data sufficient to assess the incidence of ARIA were selected. The pooled incidences of ARIA and subgroup analyses according to agent and ApoE-4 carrier status were calculated using the DerSimonian-Liard random-effects model. The proportion of symptomatic ARIA cases was also calculated. Results In total, 19 eligible studies, including 24 cohorts, were identified and 9,429 patients were analyzed. The overall pooled incidence of ARIA-effusion (E) and ARIA-hemorrhage (H) was 6.5% and 7.8%, respectively. In the subgroup analysis, the incidence of ARIA was different according to anti-Aβ immunotherapy agent. The cohorts treated with aducanumab had a significantly higher incidence of ARIA-E and ARIA-H (30.7% and 30.0%, respectively; both P < 0.001) compared to cohorts from other drugs. In subgroup analysis according to ApoE-4 carrier status, the incidences of ARIA-E and ARIA-H were higher in the ApoE-4 carrier group than those in ApoE-4 non-carrier group, but there was no statistical significance ( ApoE-4 carrier vs. non-carrier, ARIA-E; 8.6% vs. 6.9%, P = 0.663, and ARIA-H; 10.5% vs. 6.6%, P = 0.398). The pooled proportion of asymptomatic ARIA, detected by routine scheduled MRI surveillances, was 80.4%. Conclusions The overall incidences of ARIA-E and ARIA-H were 6.5% and 7.8%, respectively, and the pooled proportion of asymptomatic ARIA was 80.4%. The cohorts treated with aducanumab showed a significantly higher incidence of ARIA-E and ARIA-H (30.7% and 30.0%) compared with other drugs.

Background: High-resolution medical images that include facial regions can be used to recognize the subject's face when reconstructing 3-dimensional (3D)-rendered images from 2-dimensional (2D) sequential images, which might constitute a risk of infringement of personal information when sharing data. According to the Health Insurance Portability and Accountability Act (HIPAA) privacy rules, full-face photographic images and any comparable image are direct identifiers and considered as protected health information. Moreover, the General Data Protection Regulation (GDPR) categorizes facial images as biometric data and stipulates that special restrictions should be placed on the processing of biometric data. Objective: This study aimed to develop software that can remove the header information from Digital Imaging and Communications in Medicine (DICOM) format files and facial features (eyes, nose, and ears) at the 2D sliced-image level to anonymize personal information in medical images. Methods: A total of 240 cranial magnetic resonance (MR) images were used to train the deep learning model (144, 48, and 48 for the training, validation, and test sets, respectively, from the Alzheimer's Disease Neuroimaging Initiative [ADNI] database). To overcome the small sample size problem, we used a data augmentation technique to create 576 images per epoch. We used attention-gated U-net for the basic structure of our deep learning model. To validate the performance of the software, we adapted an external test set comprising 100 cranial MR images from the Open Access Series of Imaging Studies (OASIS) database. Results: The facial features (eyes, nose, and ears) were successfully detected and anonymized in both test sets (48 from ADNI and 100 from OASIS). Each result was manually validated in both the 2D image plane and the 3D-rendered images. Furthermore, the ADNI test set was verified using Microsoft Azure's face recognition artificial intelligence service. By adding a user interface, we developed and distributed (via GitHub) software named "Deface program" for medical images as an open-source project. Conclusions: We developed deep learning-based software for the anonymization of MR images that distorts the eyes, nose, and ears to prevent facial identification of the subject in reconstructed 3D images. It could be used to share medical big data for secondary research while making both data providers and recipients compliant with the relevant privacy regulations.

Amyloid-related imaging abnormalities (ARIA) are key safety considerations in anti-amyloid-β (Aβ) immunotherapy. ARIA can be categorized into 2 types: ARIA characterized by edema and effusion (ARIA-E) or microhemorrhages and superficial siderosis (ARIA-H). In this study, we assessed the incidence of ARIA in phase 3 randomized controlled trials (RCTs) of anti-Aβ immunotherapy and compared the incidence among different agents and APOE ε4 carrier status. PubMed and Embase databases were searched for phase 3 RCTs of anti-Aβ immunotherapy published on or before May 23, 2024. The inclusion criteria were phase 3 trials of anti-Aβ immunotherapy for mild cognitive impairment due to Alzheimer disease (AD) or mild AD dementia, with sufficient data on ARIA-E/H. The pooled incidences of ARIA and subgroup analyses according to various agents and APOE ε4 carrier status were calculated. A sensitivity analysis excluding outliers was performed. The pooled odds ratio (OR) of ARIA-E according to the APOE ε4 carrier status was also calculated. Nine phase 3 RCT cohorts from 8 eligible studies were identified, analyzing data from 6,315 patients. The pooled incidence of ARIA-E was 9.5% (95% CI 2.8%-27.3%), and the adjusted pooled incidence of ARIA-E was 25.5% (95% CI 20.4%-31.8%) in the sensitivity analysis. The pooled incidence of symptomatic ARIA-E was 6.7% (95% CI 3.5%-12.5%) and that of severe ARIA-E was 3.5% (95% CI 13.8%-8.4%). The pooled incidence of ARIA-H was 17.8% (95% CI 11.0%-27.5%), with the incidence of superficial siderosis was 9.3% (95% CI 6.1%-13.9%). The pooled incidence of isolated ARIA-H was 8.7% (95% CI 7.6%-10.1%). Subgroup analysis showed that homozygous APOE ε4 carriers had significantly higher odds of developing ARIA-E (OR 5.6, 95% CI 3.8-8.2, p < 0.001) than noncarriers. Heterozygous APOE ε4 carriers also had significantly higher odds of developing ARIA-E (OR 1.9, 95% CI 1.5-2.4, p < 0.001) than noncarriers. Although limited by small sample size and cohort-level data, our meta-analysis shows the adjusted pooled incidence of ARIA-E was 25.5% and the pooled incidence of ARIA-H was 17.8% in the recent phase 3 RCTs of anti-Aβ immunotherapy. Homozygous APOE ε4 carriers have a 5.6-fold higher risk of developing ARIA-E than noncarriers.