Predicting brain age from T1-weighted MRI is a promising marker for understanding brain aging and its associated conditions. While deep learning models have shown success in reducing the mean absolute error (MAE) of predicted brain age, concerns about robust and accurate generalization in new data limit their clinical applicability. The large number of trainable parameters, combined with limited medical imaging training data, contributes to this challenge, often resulting in a generalization gap where there is a significant discrepancy between model performance on training data versus unseen data. In this study, we assess a deep model, SFCN-reg, based on the VGG-16 architecture, and address the generalization gap through comprehensive preprocessing, extensive data augmentation, and model regularization. Using training data from the UK Biobank, we demonstrate substantial improvements in model performance. Specifically, our approach reduces the generalization MAE by 47% (from 5.25 to 2.79 years) in the Alzheimer's Disease Neuroimaging Initiative dataset and by 12% (from 4.35 to 3.75 years) in the Australian Imaging, Biomarker and Lifestyle dataset. Furthermore, we achieve up to 13% reduction in scan-rescan error (from 0.80 to 0.70 years) while enhancing the model's robustness to registration errors. Feature importance maps highlight anatomical regions used to predict age. These results highlight the critical role of high-quality preprocessing and robust training techniques in improving accuracy and narrowing the generalization gap, both necessary steps toward the clinical use of brain age prediction models. Our study makes valuable contributions to neuroimaging research by offering a potential pathway to improve the clinical applicability of deep learning models.

Machine learning models can assist with the selection of electrode arrays required for optimal insertion angles. Cochlea implantation is a successful therapy in patients with severe to profound hearing loss. The effectiveness of a cochlea implant depends on precise insertion and positioning of electrode array within the cochlea, which is known for its variability in shape and size. Preoperative imaging like CT or MRI plays a significant role in evaluating cochlear anatomy and planning the surgical approach to optimize outcomes. In this study, preoperative and postoperative CT and CBCT data of 558 cochlea-implant patients were analyzed in terms of the influence of anatomical factors and insertion depth onto the resulting insertion angle. Machine learning models can predict insertion depths needed for optimal insertion angles, with performance improving by including cochlear dimensions in the models. A simple linear regression using just the insertion depth explained 88% of variability, whereas adding cochlear length or diameter and width further improved predictions up to 94%.

Patients with identical stages or similar tumor volumes can vary significantly in their responses to radiation therapy (RT) due to individual characteristics, making personalized RT for non-small cell lung cancer (NSCLC) challenging. This study aimed to develop a deep learning model by integrating pretreatment and midtreatment computed tomography (CT) to predict the treatment response in NSCLC patients. We retrospectively collected data from 168 NSCLC patients across 3 hospitals. Data from Shanghai General Hospital (SGH, 35 patients) and Shanxi Cancer Hospital (SCH, 93 patients) were used for model training and internal validation, while data from Linfen Central Hospital (LCH, 40 patients) were used for external validation. Deep learning, radiomics, and clinical features were extracted to establish a varying time interval long short-term memory network for response prediction. Furthermore, we derived a model-deduced personalize dose escalation (DE) for patients predicted to have suboptimal gross tumor volume regression. The area under the receiver operating characteristic curve (AUC) and predicted absolute error were used to evaluate the predictive Response Evaluation Criteria in Solid Tumors classification and the proportion of gross tumor volume residual. DE was calculated as the biological equivalent dose using an /α/β ratio of 10 Gy. The model using only pretreatment CT achieved the highest AUC of 0.762 and 0.687 in internal and external validation respectively, whereas the model integrating both pretreatment and midtreatment CT achieved AUC of 0.869 and 0.798, with predicted absolute error of 0.137 and 0.185, respectively. We performed personalized DE for 29 patients. Their original biological equivalent dose was approximately 72 Gy, within the range of 71.6 Gy to 75 Gy. DE ranged from 77.7 to 120 Gy for 29 patients, with 17 patients exceeding 100 Gy and 8 patients reaching the model's preset upper limit of 120 Gy. Combining pretreatment and midtreatment CT enhances prediction performance for RT response and offers a promising approach for personalized DE in NSCLC.

Deep learning for body composition analysis (BCA) is gaining traction in clinical research, offering rapid and automated ways to measure body features like muscle or fat volume. However, most current methods prioritize computed tomography (CT) over magnetic resonance imaging (MRI). This study presents a deep learning approach for automatic BCA using MR T2-weighted sequences. Initial BCA segmentations (10 body regions and 4 body parts) were generated by mapping CT segmentations from body and organ analysis (BOA) model to synthetic MR images created using an in-house trained CycleGAN. In total, 30 synthetic data pairs were used to train an initial nnU-Net V2 in 3D, and this preliminary model was then applied to segment 120 real T2-weighted MRI sequences from 120 patients (46% female) with a median age of 56 (interquartile range, 17.75), generating early segmentation proposals. These proposals were refined by human annotators, and nnU-Net V2 2D and 3D models were trained using 5-fold cross-validation on this optimized dataset of real MR images. Performance was evaluated using Sørensen-Dice, Surface Dice, and Hausdorff Distance metrics including 95% confidence intervals for cross-validation and ensemble models. The 3D ensemble segmentation model achieved the highest Dice scores for the body region classes: bone 0.926 (95% confidence interval [CI], 0.914-0.937), muscle 0.968 (95% CI, 0.961-0.975), subcutaneous fat 0.98 (95% CI, 0.971-0.986), nervous system 0.973 (95% CI, 0.965-0.98), thoracic cavity 0.978 (95% CI, 0.969-0.984), abdominal cavity 0.989 (95% CI, 0.986-0.991), mediastinum 0.92 (95% CI, 0.901-0.936), pericardium 0.945 (95% CI, 0.924-0.96), brain 0.966 (95% CI, 0.927-0.989), and glands 0.905 (95% CI, 0.886-0.921). Furthermore, body part 2D ensemble model reached the highest Dice scores for all labels: arms 0.952 (95% CI, 0.937-0.965), head + neck 0.965 (95% CI, 0.953-0.976), legs 0.978 (95% CI, 0.968-0.988), and torso 0.99 (95% CI, 0.988-0.991). The overall average Dice across body parts (2D = 0.971, 3D = 0.969, P = ns) and body regions (2D = 0.935, 3D = 0.955, P < 0.001) ensemble models indicates stable performance across all classes. The presented approach facilitates efficient and automated extraction of BCA parameters from T2-weighted MRI sequences, providing precise and detailed body composition information across various regions and body parts.

Double-dose contrast-enhanced brain imaging improves tumor delineation and detection of occult metastases but is limited by concerns about gadolinium-based contrast agents' effects on patients and the environment. The purpose of this study was to test the benefit of a deep learning-based contrast signal amplification in true single-dose T1-weighted (T-SD) images creating artificial double-dose (A-DD) images for metastasis detection in brain magnetic resonance imaging. In this prospective, multicenter study, a deep learning-based method originally trained on noncontrast, low-dose, and T-SD brain images was applied to T-SD images of 30 participants (mean age ± SD, 58.5 ± 11.8 years; 23 women) acquired externally between November 2022 and June 2023. Four readers with different levels of experience independently reviewed T-SD and A-DD images for metastases with 4 weeks between readings. A reference reader reviewed additionally acquired true double-dose images to determine any metastases present. Performances were compared using Mid-p McNemar tests for sensitivity and Wilcoxon signed rank tests for false-positive findings. All readers found more metastases using A-DD images. The 2 experienced neuroradiologists achieved the same level of sensitivity using T-SD images (62 of 91 metastases, 68.1%). While the increase in sensitivity using A-DD images was only descriptive for 1 of them (A-DD: 65 of 91 metastases, +3.3%, P = 0.424), the second neuroradiologist benefited significantly with a sensitivity increase of 12.1% (73 of 91 metastases, P = 0.008). The 2 less experienced readers (1 resident and 1 fellow) both found significantly more metastases on A-DD images (resident, T-SD: 61.5%, A-DD: 68.1%, P = 0.039; fellow, T-SD: 58.2%, A-DD: 70.3%, P = 0.008). They were therefore able to use A-DD images to increase their sensitivity to the neuroradiologists' initial level on regular T-SD images. False-positive findings did not differ significantly between sequences. However, readers showed descriptively more false-positive findings on A-DD images. The benefit in sensitivity particularly applied to metastases ≤5 mm (5.7%-17.3% increase in sensitivity). A-DD images can improve the detectability of brain metastases without a significant loss of precision and could therefore represent a potentially valuable addition to regular single-dose brain imaging.

Noninvasive cardiac testing with coronary computed tomography angiography (CCTA) and single-photon emission computed tomography (SPECT) are becoming alternatives to invasive angiography for the evaluation of obstructive coronary artery disease. We aimed to evaluate whether a novel artificial intelligence (AI)-assisted CCTA program is comparable to SPECT imaging for ischemic testing. CCTA images were analyzed using an artificial intelligence convolutional neural network machine-learning-based model, atherosclerosis imaging-quantitative computed tomography (AI-QCT) ISCHEMIA . A total of 183 patients (75 females and 108 males, with an average age of 60.8 years ± 12.3 years) were selected. All patients underwent AI-QCT ISCHEMIA -augmented CCTA, with 60 undergoing concurrent SPECT and 16 having invasive coronary angiograms. Eight studies were excluded from analysis due to incomplete data or coronary anomalies.  A total of 175 patients (95%) had CCTA performed, deemed acceptable for AI-QCT ISCHEMIA interpretation. Compared to invasive angiography, AI-QCT ISCHEMIA -driven CCTA showed a sensitivity of 75% and specificity of 70% for predicting coronary ischemia, versus 70% and 53%, respectively for SPECT. The negative predictive value was high for female patients when using AI-QCT ISCHEMIA compared to SPECT (91% vs. 68%, P  = 0.042). Area under the receiver operating characteristic curves were similar between both modalities (0.81 for AI-CCTA, 0.75 for SPECT, P  = 0.526). When comparing both modalities, the correlation coefficient was r  = 0.71 ( P  < 0.04). AI-powered CCTA is a viable alternative to SPECT for detecting myocardial ischemia in patients with low- to intermediate-risk coronary artery disease, with significant positive and negative correlation in results. For patients who underwent confirmatory invasive angiography, the results of AI-CCTA and SPECT imaging were comparable. Future research focusing on prospective studies involving larger and more diverse patient populations is warranted to further investigate the benefits offered by AI-driven CCTA.

Deep learning reconstruction of magnetic resonance imaging (MRI) allows to either improve image quality of accelerated sequences or to generate high-resolution data. We evaluated the interaction of conventional acceleration and Deep Resolve Boost (DRB)-based reconstruction techniques of a single-shot echo-planar imaging (ssEPI) diffusion-weighted imaging (DWI) on image quality features in cerebral 3 T brain MRI and compared it with a state-of-the-art DWI sequence. In this prospective study, 24 patients received a standard of care ssEPI DWI and 5 additional adapted ssEPI DWI sequences, 3 of those with DRB reconstruction. Qualitative analysis encompassed rating of image quality, noise, sharpness, and artifacts. Quantitative analysis compared apparent diffusion coefficient (ADC) values region-wise between the different DWI sequences. Intraclass correlations, paired sampled t test, Wilcoxon signed rank test, and weighted Cohen κ were used. Compared with the reference standard, the acquisition time was significantly improved in accelerated DWI from 75 seconds up to 50% (39 seconds; P < 0.001). All tested DRB-reconstructed sequences showed significantly improved image quality, sharpness, and reduced noise ( P < 0.001). Highest image quality was observed for the combination of conventional acceleration and DL reconstruction. In singular slices, more artifacts were observed for DRB-reconstructed sequences ( P < 0.001). While in general high consistency was found between ADC values, increasing differences in ADC values were noted with increasing acceleration and application of DRB. Falsely pathological ADCs were rarely observed near frontal poles and optic chiasm attributable to susceptibility-related artifacts due to adjacent sinuses. In this comparative study, we found that the combination of conventional acceleration and DRB reconstruction improves image quality and enables faster acquisition of ssEPI DWI. Nevertheless, a tradeoff between increased acceleration with risk of stronger artifacts and high-resolution with longer acquisition time needs to be considered, especially for application in cerebral MRI.

HCC recurrence frequently occurs after curative surgery. Histological microvascular invasion (MVI) predicts recurrence but cannot provide preoperative prognostication, whereas clinical prediction scores have variable performances. Recurr-NET, a multimodal multiphasic residual-network random survival forest deep-learning model incorporating preoperative CT and clinical parameters, was developed to predict HCC recurrence. Preoperative triphasic CT scans were retrieved from patients with resected histology-confirmed HCC from 4 centers in Hong Kong (internal cohort). The internal cohort was randomly divided in an 8:2 ratio into training and internal validation. External testing was performed in an independent cohort from Taiwan.Among 1231 patients (age 62.4y, 83.1% male, 86.8% viral hepatitis, and median follow-up 65.1mo), cumulative HCC recurrence rates at years 2 and 5 were 41.8% and 56.4%, respectively. Recurr-NET achieved excellent accuracy in predicting recurrence from years 1 to 5 (internal cohort AUROC 0.770-0.857; external AUROC 0.758-0.798), significantly outperforming MVI (internal AUROC 0.518-0.590; external AUROC 0.557-0.615) and multiple clinical risk scores (ERASL-PRE, ERASL-POST, DFT, and Shim scores) (internal AUROC 0.523-0.587, external AUROC: 0.524-0.620), respectively (all p < 0.001). Recurr-NET was superior to MVI in stratifying recurrence risks at year 2 (internal: 72.5% vs. 50.0% in MVI; external: 65.3% vs. 46.6% in MVI) and year 5 (internal: 86.4% vs. 62.5% in MVI; external: 81.4% vs. 63.8% in MVI) (all p < 0.001). Recurr-NET was also superior to MVI in stratifying liver-related and all-cause mortality (all p < 0.001). The performance of Recurr-NET remained robust in subgroup analyses. Recurr-NET accurately predicted HCC recurrence, outperforming MVI and clinical prediction scores, highlighting its potential in preoperative prognostication.

The rapid evolution of artificial intelligence, especially in large language models (LLMs), has significantly impacted various domains, including healthcare. In chest X-ray (CXR) analysis, previous studies have employed LLMs, but with limitations: either underutilizing the LLMs' capability for multitask learning or lacking clinical accuracy. This article presents M4CXR, a multimodal LLM designed to enhance CXR interpretation. The model is trained on a visual instruction-following dataset that integrates various task-specific datasets in a conversational format. As a result, the model supports multiple tasks such as medical report generation (MRG), visual grounding, and visual question answering (VQA). M4CXR achieves state-of-the-art clinical accuracy in MRG by employing a chain-of-thought (CoT) prompting strategy, in which it identifies findings in CXR images and subsequently generates corresponding reports. The model is adaptable to various MRG scenarios depending on the available inputs, such as single-image, multiimage, and multistudy contexts. In addition to MRG, M4CXR performs visual grounding at a level comparable to specialized models and demonstrates outstanding performance in VQA. Both quantitative and qualitative assessments reveal M4CXR's versatility in MRG, visual grounding, and VQA, while consistently maintaining clinical accuracy.