Radiology-Artificial Intelligence最新文献

The increasing prevalence of artificial intelligence (AI)-enabled medical devices presents significant opportunities for improving patient outcomes. However, recent studies based on public U.S. Food and Drug Administration (FDA) summaries have raised concerns about the extent of validation that such devices undergo before FDA marketing authorization and subsequent clinical deployment. Here, the authors clarify key concepts of FDA regulation and provide insights into the current standards of performance validation, focusing on radiology AI devices. The authors distinguish between two fundamentally different but often conflated concepts: validation rigor (ie, the quality and comprehensiveness of the evidence supporting a device's performance) and validation transparency (ie, the extent to which this evidence is publicly accessible). The authors begin by describing the inverse relationship between the amount of performance data contained and the transparency of specific components of an FDA submission. Drawing on FDA guidelines and on experience developing authorized AI devices, the authors outline current validation standards and present a mapping from common radiology AI device types to their typical clinical study designs. This article concludes with actionable recommendations, advocating for a balanced approach tailored to specific use cases while still enforcing certain universal standards. These measures will help ensure that AI-enabled medical devices are both rigorously evaluated and transparently reported, thereby fostering greater public trust and enhancing clinical utility.

Purpose To develop a dentate nucleus (DN) segmentation tool using deep learning applied to brain MRI-based quantitative susceptibility mapping (QSM) images. Materials and Methods Brain QSM images from healthy controls and individuals with cerebellar ataxia or multiple sclerosis were collected from nine different datasets (2016-2023) worldwide for this retrospective study (ClinicalTrials.gov identifier: NCT04349514). Manual delineation of the DN was performed by experienced raters. Automated segmentation performance was evaluated against manual reference segmentations following training with several deep learning architectures. A two-step approach was used, consisting of a localization model followed by DN segmentation. Performance metrics included intraclass correlation coefficient (ICC), Dice score, and Pearson correlation coefficient. Results The training and testing datasets comprised 328 individuals (age range, 11-64 years; 171 female individuals), including 141 healthy individuals and 187 with cerebellar ataxia or multiple sclerosis. The manual tracing protocol produced reference standards with high intrarater (average ICC, 0.91) and interrater reliability (average ICC, 0.78). Initial deep learning architecture exploration indicated that the nnU-Net framework performed best. The two-step localization plus segmentation pipeline achieved a Dice score of 0.90 ± 0.03 (SD) and 0.89 ± 0.04 for left and right DN segmentation, respectively. In external testing, the proposed algorithm outperformed the current leading automated tool (mean Dice scores for left and right DN, 0.86 ± 0.04 vs 0.57 ± 0.22 [P < .001]; 0.84 ± 0.07 vs 0.58 ± 0.24 [P < .001]). The model demonstrated generalizability across datasets unseen during the training step, with automated segmentations showing high correlation with manual annotations (left DN: r = 0.74 [P < .001]; right DN: r = 0.48 [P = .03]). Conclusion The proposed model accurately and efficiently segmented the DN from brain QSM images. The model is publicly available (https://github.com/art2mri/DentateSeg). Keywords: MR Imaging, Brain/Brain Stem, Segmentation, Convolutional Neural Network, Supervised Learning, Computer Applications-3D, Volume Analysis, Image Postprocessing ClinicalTrials.gov registration no. NCT04349514 Supplemental material is available for this article. © RSNA, 2025.

Purpose To evaluate whether features extracted by Mirai can be aligned with mammographic observations and contribute meaningfully to the prediction of breast cancer risk. Materials and Methods This retrospective study examined the correlation of 512 Mirai features with mammographic observations in terms of receptive field and anatomic location. A total of 29 374 screening examinations with mammograms (10 415 female patients; mean age at examination, 60 years ± 11 [SD]) from the EMory BrEast imaging Dataset (EMBED) (2013-2020) were used to evaluate feature importance using a feature-centric explainable artificial intelligence pipeline. Risk prediction was evaluated using only calcification features (CalcMirai) or mass features (MassMirai) against Mirai. Performance was assessed in screening and screen-negative (time to cancer, >6 months) populations using the area under the receiver operating characteristic curve (AUC). Results Eighteen calcification features and 18 mass features were selected for CalcMirai and MassMirai, respectively. Both CalcMirai and MassMirai had lower performance than Mirai in lesion detection (screening population: Mirai 1-year AUC, 0.81 [95% CI: 0.78, 0.84]; CalcMirai 1-year AUC, 0.76 [95% CI: 0.73, 0.80]; MassMirai 1-year AUC, 0.74 [95% CI: 0.71, 0.78] [P < .001]). In risk prediction, there was no evidence of a difference in performance between CalcMirai and Mirai (screen-negative population: Mirai 5-year AUC, 0.66 [95% CI: 0.63, 0.69]; CalcMirai 5-year AUC, 0.66 [95% CI: 0.64, 0.69] [P = .71]). However, MassMirai achieved lower performance than Mirai (5-year AUC, 0.57 [95% CI: 0.54, 0.60]; P < .001). Radiologist review of calcification features confirmed Mirai's use of benign calcification in risk prediction. Conclusion The explainable AI pipeline demonstrated that Mirai implicitly learned to identify mammographic lesion features, particularly calcifications, for lesion detection and risk prediction. Keywords: Breast, Mammography, Screening Supplemental material is available for this article. © The Author(s) 2025. Published by the Radiological Society of North America under a CC BY 4.0 license. See also commentary by Gichoya and Trivedi in this issue.

Purpose To develop and evaluate a deep learning-based MRI denoising method using quantitative noise distribution information obtained during image reconstruction to improve model performance and generalization. Materials and Methods This retrospective study included a training set of 2 885 236 images from 96 605 cardiac cine series acquired with 3-T MRI scanners from January 2018 to December 2020. Of these data, 95% were used for training, and 5% were used for validation. The hold-out test set included 3000 cine series, acquired in the same period. Fourteen model architectures were evaluated by instantiating each of the two backbone types with seven transformer and convolution block types. The proposed SNRAware training scheme leveraged MRI reconstruction knowledge to enhance denoising by simulating diverse synthetic datasets and providing quantitative noise distribution information. Internal testing measured performance using peak signal-to-noise ratio and structural similarity index measure, whereas external tests conducted with 1.5-T real-time cardiac cine, first-pass cardiac perfusion, brain, and spine MRI assessed generalization across various sequences, contrast agents, anatomies, and field strengths. Results SNRAware improved performance on internal tests conducted on a hold-out dataset of 3000 cine series. Models trained without reconstruction knowledge achieved the worst performance metrics. Improvement was architecture agnostic for both convolution and transformer models. However, transformer models outperformed their convolutional counterparts. Additionally, three-dimensional input tensors showed improved performance over two-dimensional images. The best-performing model from the internal testing generalized well to external samples, delivering 6.5 and 2.9 times contrast-to-noise ratio improvement for real-time cine and perfusion imaging, respectively. The model trained using only cardiac cine data generalized well to three-dimensional T1-weighted magnetization-prepared rapid gradient-echo brain and T2-weighted turbo spin-echo spine MRI acquisitions. Conclusion The SNRAware training scheme leveraged data obtained during the image reconstruction process for deep learning-based MRI denoising training, resulting in improved performance and good generalization. Keywords: MRI, Deep Learning, MRI Denoising Supplemental material is available for this article. © The Author(s) 2025. Published by the Radiological Society of North America under a CC BY 4.0 license.

Purpose To evaluate the impact of screening mammography acquisition parameters on the interpretive performance of artificial intelligence (AI) and radiologists. Materials and Methods The associations between seven mammogram acquisition parameters-mammography machine version, kilovoltage peak, x-ray exposure delivered, relative x-ray exposure, paddle size, compression force, and breast thickness-and AI and radiologist performance in interpreting two-dimensional screening mammograms acquired by a diverse health system between December 2010 and 2019 were retrospectively evaluated. The top 11 AI models and the ensemble model from the Digital Mammography Dialogue on Reverse Engineering Assessment and Methods (DREAM) Challenge were assessed. The associations between each acquisition parameter and the sensitivity and specificity of the AI models and the radiologists' interpretations were separately evaluated using generalized estimating equations-based models at the examination level, adjusted for several clinical factors. Results The dataset included 28 278 screening two-dimensional mammograms from 22 626 women (mean age ± SD, 58.5 years ± 11.5; 4913 women had multiple mammograms). Of these, 324 examinations resulted in a breast cancer diagnosis within 1 year. The acquisition parameters were significantly associated with the performance of both AI and radiologists, with absolute effect sizes reaching 10% for sensitivity and 5% for specificity; however, the associations differed between AI and radiologists for several parameters. Increased exposure delivered reduced the specificity for the ensemble AI (-4.5% per 1 SD increase; P < .001) but not radiologists (P = .44). Increased compression force reduced the specificity for radiologists (-1.3% per 1 SD increase; P < .001) but not for AI (P = .60). Conclusion Screening mammography acquisition parameters impacted the performance of both AI and radiologists, with some parameters impacting performance differently. Keywords: AI Robustness, Mammography, Medical Physics Supplemental material is available for this article. © RSNA, 2025 See also commentary by Lee and Bae in this issue.

Purpose To develop a three-stage, age- and modality-conditioned framework to synthesize longitudinal infant brain MRI scans and account for rapid structural and contrast changes during early brain development. Materials and Methods This retrospective study utilized T1- and T2-weighted MRI scans (848 in total) from 139 infants in the Baby Connectome Project, collected between September 2016 and May 2020. The framework models three critical image cues related: volumetric expansion, cortical folding, and myelination, predicting missing time points with age and modality as predictive factors. The method was compared with LGAN, CounterSyn, and a diffusion-based approach using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and the Dice similarity coefficient (DSC). Results The framework was trained on 119 participants (mean age ± SD, 11.25 months ± 6.16; 60 female and 59 male infants) and tested on 20 participants (mean age, 12.98 months ± 6.59; 11 female and nine male infants). For T1-weighted images, PSNRs were 25.44 ± 1.95 and 26.93 ± 2.50 for forward and backward MRI synthesis, respectively, and SSIMs were 0.87 ± 0.03 and 0.90 ± 0.02, respectively. For T2-weighted images, PSNRs were 26.35 ± 2.30 and 26.40 ± 2.56, respectively, with SSIMs of 0.87 ± 0.03 and 0.89 ± 0.02, respectively, showing significant outperformance compared with competing methods (P < .001). The framework also excelled in tissue segmentation (P < .001) and cortical reconstruction, achieving a DSC of 0.85 for gray matter and 0.86 for white matter, with intraclass correlation coefficients exceeding 0.8 in most cortical regions. Conclusion The proposed three-stage framework effectively synthesized age-specific infant brain MRI scans, outperforming competing methods in image quality and tissue segmentation and with strong performance in cortical reconstruction, demonstrating potential for developmental modeling and longitudinal analyses. Keywords: Pediatrics, Brain, Brain Stem, MRI, Infant Brain MRI Supplemental material is available for this article. © RSNA, 2025 See also commentary by Chaudhari and Rauschecker in this issue.