Zechen Zhou

Background: Deep learning (DL)‐based image enhancement is widely used to improve suboptimal medical imaging. Whether it also benefits diagnostic‐quality MRI in downstream task performance and data‐efficiency remains unclear. Purpose: To investigate the impact of DL‐based enhancement applied to diagnostic quality structural MRI for Alzheimer's disease (AD) classification. Study Type: Retrospective. Population: A total of 2293 brain MRI scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI) were split into training ( n = 1605), validation ( n = 229), and internal test ( n = 459) sets. Two hundred and seventy scans from the National Alzheimer's Coordinating Center (NACC) served as an external test set. Field Strength/Sequence: 1.5 T; 3D T1‐weighted gradient‐echo. Assessment: Each scan was enhanced by SubtleHD (SHD), an FDA‐cleared DL‐based MR enhancement tool. ResNet34 and DenseNet121 were trained on standard‐of‐care (SOC) and SHD‐enhanced images to classify subjects as cognitively normal, mild cognitive impairment, or AD and evaluated by accuracy and macro–area under the receiver operating characteristic curve (macro‐AUC). Data efficiency was assessed by retraining on stratified training subsets (50%–100%). Statistical Tests: McNemar test for accuracy and DeLong test for macro‐AUC in three‐class one‐versus‐rest setting ( p < 0.05). Results: SHD enhancement increased ResNet34 accuracy from 85.2% to 88.7% and macro‐AUC from 0.951 to 0.968 (both significant), and DenseNet121 accuracy from 90.2% to 92.2% ( p = 0.18) and macro‐AUC from 0.978 to 0.982 ( p = 0.29). Models trained on 70% of SHD‐enhanced dataset matched those trained on the full SOC dataset (accuracy: 85.9%, macro‐AUC: 0.942), indicating improved data efficiency with enhancement. In NACC, the SOC‐trained model achieved accuracy of 49.2% and macro‐AUC of 0.679 versus 63.0% and 0.819 for the SHD‐trained model (both significant); the SHD‐trained model retained an advantage on unenhanced NACC images (macro‐AUC: 0.772). Data Conclusion: DL‐based enhancement of diagnostic‐quality MRI improves downstream Alzheimer's disease classification performance and reduces the amount of training data required. This suggests that conventional definitions of image quality may underestimate the information content available for machine learning. Evidence Level: 3. Technical Efficacy Stage: 2.

Background: Deep learning (DL) based image enhancement is widely used to improve suboptimal medical imaging. However, whether such techniques can extract additional diagnostically relevant information from already high quality, clinically acceptable MRI, and thereby improve downstream artificial intelligence (AI) model performance and data-efficiency, remains unclear. We investigated the impact of DL-based enhancement on AI-based Alzheimer’s disease (AD) classification from structural MRI. Methods: In this retrospective study, we included 2,293 MRI scans from the public Alzheimer’s Disease Neuroimaging Initiative dataset (training/validation/test 7:1:2), upon which multiple AI classification models were trained, first on the original images (standard-of-care [SOC]), then on the same images enhanced by SubtleHD (SHD), an FDA-cleared enhancement tool. Convolutional neural networks (ResNet34 and DenseNet121) were used to classify subjects as cognitively normal (CN), mild cognitive impairment (MCI), or AD. Model performance was assessed using accuracy and macro–area under the receiver operating characteristic curve (macro-AUC). Data efficiency was assessed by training on progressively smaller subsets of the full dataset. Findings: Models trained and evaluated on enhanced images consistently outperformed those using the original images. For ResNet34, enhancement improved classification accuracy from 85·2% to 88·7% and macro-AUC from 0·951 to 0·968 (both p<0·01). Similar gains were observed with DenseNet121. Notably, models trained on 75% of SHD-enhanced images achieved comparable performance to models trained on the full SOC dataset (accuracy 86·5%, macro-AUC 0·954), indicating improved data efficiency.Interpretations: DL-based enhancement of MRI image quality, even when images are already considered to be of diagnostic quality, improves downstream AI classification performance and reduces the amount of training data required. This has profound implications for radiologic imaging, suggesting that conventional definitions of image quality may underestimate the information content available for machine learning. DL-based enhancement strategies could also enable more efficient development and broader deployment of AI models in neuroimaging, particularly in data-limited settings.

Purpose: To propose a highly time-efficient imaging technique namedimproved simultaneous noncontrast angiography and intraplaque hemorrhage(iSNAP) for simultaneous assessment of lumen, vessel wall, and blood flow in intracranial arteries. Methods: iSNAP consists of pulsed arterial spin labeling preparations and 3D golden angle radial acquisition. Images were reconstructed by k-space weighted image contrast (KWIC) method with optimized data-sharing strategies. Dynamic MRA for blood flow assessment was obtained from iSNAP by reconstruction at multiple inversion times and image subtraction, static MRA by both image subtraction approach and phase-sensitive inversion recovery technique, and vessel wall images by both reconstruction at zero-crossing time-point of blood and phase-sensitive inversion recovery. A T-1-weighted brain MRI was also reconstructed from iSNAP. Preliminary comparison of iSNAP against the dedicated dynamic MRA sequence 4D-TRANCE, MRA/vessel wall imaging sequence SNAP, and vessel wall imaging sequence T-1-weighted VISTA was performed in healthy volunteers and patients. Results: iSNAP has whole-brain coverage and takes similar to 6.5 min. The dedicated reconstruction strategies are feasible for each iSNAP image contrast and beneficial for image SNR. iSNAP-dynamic MRA yields similar dynamic flow information as 4D-TRANCE and allows more flexible temporal resolution. The 2 types of iSNAP static MRA images complement each other in characterizing both proximal large arteries and distal small arteries. Depiction of vessel wall lesions in iSNAP vessel wall images is better than SNAP and may be similar to T-1-weighted VISTA, although the images are slightly blurred. Conclusion: iSNAP provides a time-efficient evaluation of intracranial arteries and may have great potential for comprehensive assessment of intracranial vascular conditions using a single sequence.

Purpose To develop a self‐calibrating approach for the estimation of wave point spread function (PSF) and coil sensitivities from the subsampled wave‐encoded k‐space, and evaluate its performance for wave‐encoded 3D turbo spin echo (TSE) imaging. Methods A low rank subspace parametric model was demonstrated in simulation to improve the representation for practical wave encoding k‐space trajectories with aperiodicity, and an autofocus metric for the entire imaging volume was used to calibrate the wave PSF in a 2‐stage manner from coarse to refined estimation. The coil sensitivities can be extracted from the shifted central region of wave PSF corrected subsampled k‐space, and further used with wave PSF for wave‐encoded parallel imaging (PI) reconstruction. The wave encoding gradients were integrated into the 3D TSE sequence considering eddy current reduction aspects and maintaining of the Carr‐Purcell‐Meiboom‐Gill condition. Phantom and in vivo brain experiments were performed to evaluate the accuracy of wave PSF self‐calibration and to compare the PI reconstruction performance between wave and Cartesian encoding scheme. Results The self‐calibrated wave PSF, estimated from different k‐space undersampling patterns can robustly correct the wave encoding induced image artifacts given sufficient central autocalibration data. The self‐calibrating wave‐encoded PI reconstruction has demonstrated its improved performance in reduced aliasing artifacts and noise amplification in comparison to the Cartesian‐encoded PI reconstruction results for 3D TSE imaging. Conclusion The proposed self‐calibrating wave‐encoded method allows robust calibration of wave PSF and coil sensitivities from the subsampled k‐space, and improves the overall image quality for accelerated 3D TSE imaging.

Objective To develop a 3D multi-contrast IVW protocol with 0.5-mm isotropic resolution and a scan time of 5 min per sequence. Materials and methods Pre-contrast T1w VISTA, DANTE prepared PDw VISTA, SNAP, and post-contrast T1w VISTA were accelerated using cartesian undersampling with target ordering method (CUSTOM) and self-supporting tailored k-space estimation for parallel imaging reconstruction (STEP). CUSTOM + STEP IVW was compared to full-sample IVW, SENSE-accelerated IVW, and CUSTOM + zero-filled Fourier reconstruction in normal volunteers and subjects with intracranial atherosclerotic disease (ICAD). Image quality, vessel delineation, CSF suppression, and blood suppression were compared. Results CUSTOM + STEP vessel wall delineation was comparable to full-sample IVW and better than SENSE IVW for vessel wall delineation on T1w VISTA and luminal contrast on SNAP. Average image quality and wall depiction were significantly improved using STEP reconstruction compared with zero-filled Fourier reconstruction, with no significant difference in CSF or blood suppression. Conclusions CUSTOM + STEP allowed multi-contrast 3D 0.5-mm isotropic IVW within 30 min. Although some quantitative and qualitative scores for CUSTOM − STEP were lower than fully sampled IVW, CUSTOM + STEP provided comparable vessel wall delineation as full-sample IVW and was superior to SENSE. CUSTOM + STEP IVW was well tolerated by patients and showed good delineation of ICAD plaque.