Search

Background: Contrastive learning, a successful form of representational learning, has shown promising results in pretraining deep learning (DL) models for downstream tasks. When working with limited annotation data, as in medical image segmentation tasks, learning domain‐specific local representations can further improve the performance of DL models. Purpose: In this work, we extend the contrastive learning framework to utilize domain‐specific contrast information from unlabeled Magnetic Resonance (MR) images to improve the performance of downstream MR image segmentation tasks in the presence of limited labeled data. Methods: The contrast in MR images is controlled by underlying tissue properties (e.g., T1 or T2) and image acquisition parameters. We hypothesize that learning to discriminate local representations based on underlying tissue properties should improve subsequent segmentation tasks on MR images. We propose a novel constrained contrastive learning (CCL) strategy that uses tissue‐specific information via a constraint map to define positive and negative local neighborhoods for contrastive learning, embedding this information in the representational space during pretraining. For a given MR contrast image, the proposed strategy uses local signal characteristics (constraint map) across a set of related multi‐contrast MR images as a surrogate for underlying tissue information. We demonstrate the utility of the approach for downstream: (1) multi‐organ segmentation tasks in T2‐weighted images where a DL model learns T2 information with constraint maps from a set of 2D multi‐echo T2‐weighted images (n = 101) and (2) tumor segmentation tasks in multi‐parametric images from the public brain tumor segmentation (BraTS) (n = 80) dataset where DL models learn T1 and T2 information from multi‐parametric BraTS images. Performance is evaluated on downstream multi‐label segmentation tasks with limited data in (1) T2‐weighted images of the abdomen from an in‐house Radial‐T2 (Train/Test = 30/20), (2) public Cartesian‐T2 (Train/Test = 6/12) dataset, and (3) multi‐parametric MR images from the public brain tumor segmentation dataset (BraTS) (Train/Test = 40/50). The performance of the proposed CCL strategy is compared to state‐of‐the‐art self‐supervised contrastive learning techniques. In each task, a model is also trained using all available labeled data for supervised baseline performance. Results: The proposed CCL strategy consistently yielded improved Dice scores, Precision, and Recall metrics, and reduced HD95 values across all segmentation tasks. We also observed performance comparable to the baseline with reduced annotation effort. The t‐SNE visualization of features for T2‐weighted images demonstrates its ability to embed T2 information in the representational space. On the BraTS dataset, we also observed that using an appropriate multi‐contrast space to learn T1+T2, T1, or T2 information during pretraining further improved the performance of tumor segmentation tasks. Conclusions: Learning to embed tissue‐specific information that controls MR image contrast with the proposed constrained contrastive learning improved the performance of DL models on subsequent segmentation tasks compared to conventional self‐supervised contrastive learning techniques. The use of such domain‐specific local representations could help understand, improve performance, and mitigate the scarcity of labeled data in MR image segmentation tasks. [ABSTRACT FROM AUTHOR]

Background: T2 mapping is of great interest in abdominal imaging but current methods are limited by low resolution, slice coverage, motion sensitivity, or lengthy acquisitions. Purpose Develop a radial turbo spin‐echo technique with refocusing variable flip angles (RADTSE‐VFA) for high spatiotemporal T2 mapping and efficient slice coverage within a breath‐hold and compare to the constant flip angle counterpart (RADTSE‐CFA). Study Type: Prospective technical efficacy. Subjects: Testing performed on agarose phantoms and 12 patients. Focal liver lesion classification tested on malignant (N = 24) and benign (N = 11) lesions. Field Strength/Sequence: 1.5 T/RADTSE‐VFA, RADTSE‐CFA. Assessment A constrained objective function was used to optimize the refocusing flip angles. Phantom and/or in vivo data were used to assess relative contrast, T2 estimation, specific absorption rate (SAR), and focal liver lesion classification. Statistical Tests: t‐Tests or Mann–Whitney Rank Sum tests were used. Results: Phantom data did not show significant differences in mean relative contrast (P = 0.10) and T2 accuracy (P = 0.99) between RADTSE‐VFA and RADTSE‐CFA. Adding noise caused T2 overestimation predominantly for RADTSE‐CFA and low T2 values. In vivo results did not show significant differences in mean spleen‐to‐liver (P = 0.62) and kidney‐to‐liver (P = 0.49) relative contrast between RADTSE‐VFA and RADTSE‐CFA. Mean T2 values were not significantly different between the two techniques for spleen (T2VFA = 109.2 ± 12.3 msec; T2CFA = 110.7 ± 11.1 msec; P = 0.78) and kidney‐medulla (T2VFA = 113.0 ± 8.7 msec; T2CFA = 114.0 ± 8.6 msec; P = 0.79). Liver T2 was significantly higher for RADTSE‐CFA (T2VFA = 52.6 ± 6.6 msec; T2CFA = 60.4 ± 8.0 msec) consistent with T2 overestimation in the phantom study. Focal liver lesion classification had comparable T2 distributions for RADTSE‐VFA and RADTSE‐CFA for malignancies (P = 1.0) and benign lesions (P = 0.39). RADTSE‐VFA had significantly lower SAR than RADTSE‐CFA increasing slice coverage by 1.5. Data Conclusion: RADTSE‐VFA provided noise‐robust T2 estimation compared to the constant flip angle counterpart while generating T2‐weighted images with comparable contrast. The VFA scheme minimized SAR improving slice efficiency for breath‐hold imaging. Level of Evidence: 2 Technical Efficacy Stage: 1 [ABSTRACT FROM AUTHOR]

The article discusses the use of quantitative estimation of longitudinal relaxation time T1 in cardiac MRI to visualize and quantify changes in the myocardium. Traditional methods for T1 mapping involve long data acquisition times, which may not be suitable for all patients. Recent deep learning-based techniques have shown promise in reducing acquisition time while maintaining precision. The article presents a study that demonstrates the generalizability of a deep learning model called MyoMapNet to multi-institutional data from different scanner manufacturers and field strengths. The results show strong correlation between native and postcontrast T1 estimates from the model and those from traditional methods. The article concludes by suggesting future research directions to improve the adaptability of deep learning models in broader clinical environments. [Extracted from the article]