A review on medical imaging synthesis using deep learning and its clinical applications

Artificial intelligence-based motion tracking in cancer radiotherapy: A review

Radiotherapy aims to deliver a prescribed dose to the tumor while sparing neighboring organs at risk (OARs). Increasingly complex treatment techniques such as volumetric modulated arc therapy (VMAT), stereotactic radiosurgery (SRS), stereotactic body radiotherapy (SBRT), and proton therapy have been developed to deliver doses more precisely to the target. While such technologies have improved dose delivery, the implementation of intra-fraction motion management to verify tumor position at the time of treatment has become increasingly relevant. Artificial intelligence (AI) has recently demonstrated great potential for real-time tracking of tumors during treatment. However, AI-based motion management faces several challenges, including bias in training data, poor transparency, difficult data collection, complex workflows and quality assurance, and limited sample sizes. This review presents the AI algorithms used for chest, abdomen, and pelvic tumor motion management/tracking for radiotherapy and provides a literature summary on the topic. We will also discuss the limitations of these AI-based studies and propose potential improvements.

CBCT-based synthetic CT image generation using a diffusion model for CBCT-guided lung radiotherapy

BACKGROUND

Although cone beam computed tomography (CBCT) has lower resolution compared to planning CTs (pCT), its lower dose, higher high-contrast resolution, and shorter scanning time support its widespread use in clinical applications, especially in ensuring accurate patient positioning during the image-guided radiation therapy (IGRT) process.

PURPOSE

While CBCT is critical to IGRT, CBCT image quality can be compromised by severe stripe and scattering artifacts. Tumor movement secondary to respiratory motion also decreases CBCT resolution. In order to improve the image quality of CBCT, we propose a Lung Diffusion Model (L-DM) framework.

METHODS

Our proposed algorithm is based on a conditional diffusion model trained on pCT and deformed CBCT (dCBCT) image pairs to synthesize lung CT images from dCBCT images and benefit CBCT-based radiotherapy. dCBCT images were used as the constraint for the L-DM. The image quality and Hounsfield unit (HU) values of the synthetic CTs (sCT) images generated by the proposed L-DM were compared to three selected mainstream generation models.

RESULTS

We verified our model in both an institutional lung cancer dataset and a selected public dataset. Our L-DM showed significant improvement in the four metrics of mean absolute error (MAE), peak signal-to-noise ratio (PSNR), normalized cross-correlation (NCC), and structural similarity index measure (SSIM). In our institutional dataset, our proposed L-DM decreased the MAE from 101.47 to 37.87 HU and increased the PSNR from 24.97 to 29.89 dB, the NCC from 0.81 to 0.97, and the SSIM from 0.80 to 0.93. In the public dataset, our proposed L-DM decreased the MAE from 173.65 to 58.95 HU, while increasing the PSNR, NCC, and SSIM from 13.07 to 24.05 dB, 0.68 to 0.94, and 0.41 to 0.88, respectively.

CONCLUSIONS

The proposed L-DM significantly improved sCT image quality compared to the pre-correction CBCT and three mainstream generative models. Our model can benefit CBCT-based IGRT and other potential clinical applications as it increases the HU accuracy and decreases the artifacts from input CBCT images.

Using a patient-specific diffusion model to generate CBCT-based synthetic CTs for CBCT-guided adaptive radiotherapy

BACKGROUND

Cone beam computed tomography (CBCT) can be used to evaluate the inter-fraction anatomical changes during the entire course for image-guided radiotherapy (IGRT). However, CBCT artifacts from various sources restrict the full application of CBCT-guided adaptive radiation therapy (ART).

PURPOSE

Inter-fraction anatomical changes during ART, including variations in tumor size and normal tissue anatomy, can affect radiation therapy (RT) efficacy. Acquiring high-quality CBCT images that accurately capture patient- and fraction-specific (PFS) anatomical changes is crucial for successful IGRT.

METHODS

To enhance CBCT image quality, we proposed PFS lung diffusion models (PFS-LDMs). The proposed PFS models use a pre-trained general lung diffusion model (GLDM) as a baseline, which is trained on historical deformed CBCT (dCBCT)-planning CT (pCT) paired data. For a given patient, a new PFS model is fine-tuned on a CBCT-deformed pCT (dpCT) pair after each fraction to learn the PFS knowledge for generating personalized synthetic CT (sCT) with quality comparable to pCT or dpCT. The learned PFS knowledge is the specific mapping relationships, including personal inter-fraction anatomical changes between personalized CBCT-dpCT pairs. The PFS-LDMs were evaluated on an institutional lung cancer dataset, quantified by mean absolute error (MAE), peak signal-to-noise ratio (PSNR), normalized cross-correlation (NCC), and structural similarity index measure (SSIM) metrics. We also compared our PFS-LDMs with a mainstream GAN-based model, demonstrating that our PFS fine-tuning strategy could be applied to existing generative models.

RESULTS

Our models showed remarkable improvements across all four evaluation metrics. The proposed PFS-LDMs outperformed the GLDM, demonstrating the effectiveness of our proposed fine-tuning strategy. The PFS model fine-tuned with CBCT images from four prior fractions, reduced the MAE from 103.95 to 15.96 Hounsfield units (HU), and increased the mean PSNR, NCC, and SSIM from 25.36 dB to 33.57 dB, 0.77 to 0.98, and 0.75 to 0.97, respectively. Applying our PFS fine-tuning strategy to a Cycle GAN model also showed improvements, with all four fine-tuned PFS Cycle GAN (PFS-CG) models outperforming the general Cycle GAN model. Overall, our proposed PFS fine-tuning strategy improved CBCT image quality compared to both the pre-correction and non-fine-tuned general models, with our proposed PFS-LDMs yielding better performance than the GAN-based model across all metrics.

CONCLUSIONS

Our proposed PFS-LDMs significantly improve CBCT image quality with increased HU accuracy and fewer artifacts, thus better capturing inter-fraction anatomical changes. This lays the groundwork for enabling CBCT-based ART, which could enhance clinical efficiency and achieve personalized high-precision treatment by accounting for inter-fraction anatomical changes.

CCSI: Continual Class-Specific Impression for data-free class incremental learning

In real-world clinical settings, traditional deep learning-based classification methods struggle with diagnosing newly introduced disease types because they require samples from all disease classes for offline training. Class incremental learning offers a promising solution by adapting a deep network trained on specific disease classes to handle new diseases. However, catastrophic forgetting occurs, decreasing the performance of earlier classes when adapting the model to new data. Prior proposed methodologies to overcome this require perpetual storage of previous samples, posing potential practical concerns regarding privacy and storage regulations in healthcare. To this end, we propose a novel data-free class incremental learning framework that utilizes data synthesis on learned classes instead of data storage from previous classes. Our key contributions include acquiring synthetic data known as Continual Class-Specific Impression (CCSI) for previously inaccessible trained classes and presenting a methodology to effectively utilize this data for updating networks when introducing new classes. We obtain CCSI by employing data inversion over gradients of the trained classification model on previous classes starting from the mean image of each class inspired by common landmarks shared among medical images and utilizing continual normalization layers statistics as a regularizer in this pixel-wise optimization process. Subsequently, we update the network by combining the synthesized data with new class data and incorporate several losses, including an intra-domain contrastive loss to generalize the deep network trained on the synthesized data to real data, a margin loss to increase separation among previous classes and new ones, and a cosine-normalized cross-entropy loss to alleviate the adverse effects of imbalanced distributions in training data. Extensive experiments show that the proposed framework achieves state-of-the-art performance on four of the public MedMNIST datasets and in-house echocardiography cine series, with an improvement in classification accuracy of up to 51% compared to baseline data-free methods. Our code is available at https://github.com/ubc-tea/Continual-Impression-CCSI.

Reducing annotating load: Active learning with synthetic images in surgical instrument segmentation

Accurate instrument segmentation in the endoscopic vision of minimally invasive surgery is challenging due to complex instruments and environments. Deep learning techniques have shown competitive performance in recent years. However, deep learning usually requires a large amount of labeled data to achieve accurate prediction, which poses a significant workload. To alleviate this workload, we propose an active learning-based framework to generate synthetic images for efficient neural network training. In each active learning iteration, a small number of informative unlabeled images are first queried by active learning and manually labeled. Next, synthetic images are generated based on these selected images. The instruments and backgrounds are cropped out and randomly combined with blending and fusion near the boundary. The proposed method leverages the advantage of both active learning and synthetic images. The effectiveness of the proposed method is validated on two sinus surgery datasets and one intraabdominal surgery dataset. The results indicate a considerable performance improvement, especially when the size of the annotated dataset is small. All the code is open-sourced at: https://github.com/HaonanPeng/active_syn_generator.

Synthetic Low-Energy Monochromatic Image Generation in Single-Energy Computed Tomography System Using a Transformer-Based Deep Learning Model

While dual-energy computed tomography (DECT) technology introduces energy-specific information in clinical practice, single-energy CT (SECT) is predominantly used, limiting the number of people who can benefit from DECT. This study proposed a novel method to generate synthetic low-energy virtual monochromatic images at 50 keV (sVMI50keV) from SECT images using a transformer-based deep learning model, SwinUNETR. Data were obtained from 85 patients who underwent head and neck radiotherapy. Among these, the model was built using data from 70 patients for whom only DECT images were available. The remaining 15 patients, for whom both DECT and SECT images were available, were used to predict from the actual SECT images. We used the SwinUNETR model to generate sVMI50keV. The image quality was evaluated, and the results were compared with those of the convolutional neural network-based model, Unet. The mean absolute errors from the true VMI50keV were 36.5 ± 4.9 and 33.0 ± 4.4 Hounsfield units for Unet and SwinUNETR, respectively. SwinUNETR yielded smaller errors in tissue attenuation values compared with those of Unet. The contrast changes in sVMI50keV generated by SwinUNETR from SECT were closer to those of DECT-derived VMI50keV than the contrast changes in Unet-generated sVMI50keV. This study demonstrated the potential of transformer-based models for generating synthetic low-energy VMIs from SECT images, thereby improving the image quality of head and neck cancer imaging. It provides a practical and feasible solution to obtain low-energy VMIs from SECT data that can benefit a large number of facilities and patients without access to DECT technology.

DuDoCFNet: Dual-Domain Coarse-to-Fine Progressive Network for Simultaneous Denoising, Limited-View Reconstruction, and Attenuation Correction of Cardiac SPECT

Single-Photon Emission Computed Tomography (SPECT) is widely applied for the diagnosis of coronary artery diseases. Low-dose (LD) SPECT aims to minimize radiation exposure but leads to increased image noise. Limited-view (LV) SPECT, such as the latest GE MyoSPECT ES system, enables accelerated scanning and reduces hardware expenses but degrades reconstruction accuracy. Additionally, Computed Tomography (CT) is commonly used to derive attenuation maps ( μ -maps) for attenuation correction (AC) of cardiac SPECT, but it will introduce additional radiation exposure and SPECT-CT misalignments. Although various methods have been developed to solely focus on LD denoising, LV reconstruction, or CT-free AC in SPECT, the solution for simultaneously addressing these tasks remains challenging and under-explored. Furthermore, it is essential to explore the potential of fusing cross-domain and cross-modality information across these interrelated tasks to further enhance the accuracy of each task. Thus, we propose a Dual-Domain Coarse-to-Fine Progressive Network (DuDoCFNet), a multi-task learning method for simultaneous LD denoising, LV reconstruction, and CT-free μ -map generation of cardiac SPECT. Paired dual-domain networks in DuDoCFNet are cascaded using a multi-layer fusion mechanism for cross-domain and cross-modality feature fusion. Two-stage progressive learning strategies are applied in both projection and image domains to achieve coarse-to-fine estimations of SPECT projections and CT-derived μ -maps. Our experiments demonstrate DuDoCFNet's superior accuracy in estimating projections, generating μ -maps, and AC reconstructions compared to existing single- or multi-task learning methods, under various iterations and LD levels. The source code of this work is available at https://github.com/XiongchaoChen/DuDoCFNet-MultiTask.

Deep learning for the harmonization of structural MRI scans: a survey

Medical imaging datasets for research are frequently collected from multiple imaging centers using different scanners, protocols, and settings. These variations affect data consistency and compatibility across different sources. Image harmonization is a critical step to mitigate the effects of factors like inherent differences between various vendors, hardware upgrades, protocol changes, and scanner calibration drift, as well as to ensure consistent data for medical image processing techniques. Given the critical importance and widespread relevance of this issue, a vast array of image harmonization methodologies have emerged, with deep learning-based approaches driving substantial advancements in recent times. The goal of this review paper is to examine the latest deep learning techniques employed for image harmonization by analyzing cutting-edge architectural approaches in the field of medical image harmonization, evaluating both their strengths and limitations. This paper begins by providing a comprehensive fundamental overview of image harmonization strategies, covering three critical aspects: established imaging datasets, commonly used evaluation metrics, and characteristics of different scanners. Subsequently, this paper analyzes recent structural MRI (Magnetic Resonance Imaging) harmonization techniques based on network architecture, network learning algorithm, network supervision strategy, and network output. The underlying architectures include U-Net, Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), flow-based generative models, transformer-based approaches, as well as custom-designed network architectures. This paper investigates the effectiveness of Disentangled Representation Learning (DRL) as a pivotal learning algorithm in harmonization. Lastly, the review highlights the primary limitations in harmonization techniques, specifically the lack of comprehensive quantitative comparisons across different methods. The overall aim of this review is to serve as a guide for researchers and practitioners to select appropriate architectures based on their specific conditions and requirements. It also aims to foster discussions around ongoing challenges in the field and shed light on promising future research directions with the potential for significant advancements.

On-board synthetic 4D MRI generation from 4D CBCT for radiotherapy of abdominal tumors: A feasibility study

BACKGROUND

Magnetic resonance-guided radiotherapy with an MR-guided LINAC represents potential clinical benefits in abdominal treatments due to the superior soft-tissue contrast compared to kV-based images in conventional treatment units. However, due to the high cost associated with this technology, only a few centers have access to it. As an alternative, synthetic 4D MRI generation based on artificial intelligence methods could be implemented. Nevertheless, appropriate MRI texture generation from CT images might be challenging and prone to hallucinations, compromising motion accuracy.

PURPOSE

To evaluate the feasibility of on-board synthetic motion-resolved 4D MRI generation from prior 4D MRI, on-board 4D cone beam CT (CBCT) images, motion modeling information, and deep learning models using the digital anthropomorphic phantom XCAT.

METHODS

The synthetic 4D MRI corresponds to phases from on-board 4D CBCT. Each synthetic MRI volume in the 4D MRI was generated by warping a reference 3D MRI (MRIref, end of expiration phase from a prior 4D MRI) with a deformation field map (DFM) determined by (I) the eigenvectors from the principal component analysis (PCA) motion-modeling of the prior 4D MRI, and (II) the corresponding eigenvalues predicted by a convolutional neural network (CNN) model using the on-board 4D CBCT images as input. The CNN was trained with 1000 deformations of one reference CT (CTref, same conditions as MRIref) generated by applying 1000 DFMs computed by randomly sampling the original eigenvalues from the prior 4D MRI PCA model. The evaluation metrics for the CNN model were root-mean-square error (RMSE) and mean absolute error (MAE). Finally, different on-board 4D-MRI generation scenarios were assessed by changing the respiratory period, the amplitude of the diaphragm, and the chest wall motion of the 4D CBCT using normalized root-mean-square error (nRMSE) and structural similarity index measure (SSIM) for image-based evaluation, and volume dice coefficient (VDC), volume percent difference (VPD), and center-of-mass shift (COMS) for contour-based evaluation of liver and target volumes.

RESULTS

The RMSE and MAE values of the CNN model reported 0.012 ± 0.001 and 0.010 ± 0.001, respectively for the first eigenvalue predictions. SSIM and nRMSE were 0.96 ± 0.06 and 0.22 ± 0.08, respectively. VDC, VPD, and COMS were 0.92 ± 0.06, 3.08 ± 3.73 %, and 2.3 ± 2.1 mm, respectively, for the target volume. The more challenging synthetic 4D-MRI generation scenario was for one 4D-CBCT with increased chest wall motion amplitude, reporting SSIM and nRMSE of 0.82 and 0.51, respectively.

CONCLUSIONS

On-board synthetic 4D-MRI generation based on predicting actual treatment deformation from on-board 4D-CBCT represents a method that can potentially improve the treatment-setup localization in abdominal radiotherapy treatments with a conventional kV-based LINAC.

Clinical feasibility of deep learning based synthetic contrast enhanced abdominal CT in patients undergoing non enhanced CT scans

Our objective was to develop and evaluate the clinical feasibility of deep-learning-based synthetic contrast-enhanced computed tomography (DL-SynCCT) in patients designated for nonenhanced CT (NECT). We proposed a weakly supervised learning with the utilization of virtual non-contrast CT (VNC) for the development of DL-SynCCT. Training and internal validations were performed with 2202 pairs of retrospectively collected contrast-enhanced CT (CECT) images with the corresponding VNC images acquired from dual-energy CT. Clinical validation was performed using an external validation set including 398 patients designated for true nonenhanced CT (NECT), from multiple vendors at three institutes. Detection of lesions was performed by three radiologists with only NECT in the first session and an additionally provided DL-SynCCT in the second session. The mean peak signal-to-noise ratio (PSNR) and structural similarity index map (SSIM) of the DL-SynCCT compared to CECT were 43.25 ± 0.41 and 0.92 ± 0.01, respectively. With DL-SynCCT, the pooled sensitivity for lesion detection (72.0% to 76.4%, P < 0.001) and level of diagnostic confidence (3.0 to 3.6, P < 0.001) significantly increased. In conclusion, DL-SynCCT generated by weakly supervised learning showed significant benefit in terms of sensitivity in detecting abnormal findings when added to NECT in patients designated for nonenhanced CT scans.

DiffuseRT: predicting likely anatomical deformations of patients undergoing radiotherapy

Objective. Predicting potential deformations of patients can improve radiotherapy treatment planning. Here, we introduce new deep-learning models that predict likely anatomical changes during radiotherapy for head and neck cancer patients.Approach. Denoising diffusion probabilistic models (DDPMs) were developed to generate fraction-specific anatomical changes based on a reference cone-beam CT (CBCT), the fraction number and the dose distribution delivered. Three distinct DDPMs were developed: (1) theimage modelwas trained to directly generate likely future CBCTs, (2) the deformable vector field (DVF) model was trained to generate DVFs that deform a reference CBCT and (3) thehybrid modelwas trained similarly to the DVF model, but without relying on an external deformable registration algorithm. The models were trained on 9 patients with longitudinal CBCT images (224 CBCTs) and evaluated on 5 patients (152 CBCTs).Results. The generated images mainly exhibited random positioning shifts and small anatomical changes for early fractions. For later fractions, all models predicted weight losses in accordance with the training data. The distributions of volume and position changes of the body, esophagus, and parotids generated with the image and hybrid models were more similar to the ground truth distribution than the DVF model, evident from the lower Wasserstein distance achieved with the image (0.33) and hybrid model (0.30) compared to the DVF model (0.36). Generating several images for the same fraction did not yield the expected variability since the ground truth anatomical changes were only in 76% of the fractions within the 95% bounds predicted with the best model. Using the generated images for robust optimization of simplified proton therapy plans improved the worst-case clinical target volume V95 with 7% compared to optimizing with 3 mm set-up robustness while maintaining a similar integral dose.Significance. The newly developed DDPMs generate distributions similar to the real anatomical changes and have the potential to be used for robust anatomical optimization.

Uncertainty quantification via localized gradients for deep learning-based medical image assessments

Objective.Deep learning models that aid in medical image assessment tasks must be both accurate and reliable to be deployed within clinical settings. While deep learning models have been shown to be highly accurate across a variety of tasks, measures that indicate the reliability of these models are less established. Increasingly, uncertainty quantification (UQ) methods are being introduced to inform users on the reliability of model outputs. However, most existing methods cannot be augmented to previously validated models because they are not post hoc, and they change a model's output. In this work, we overcome these limitations by introducing a novel post hoc UQ method, termedLocal Gradients UQ, and demonstrate its utility for deep learning-based metastatic disease delineation.Approach.This method leverages a trained model's localized gradient space to assess sensitivities to trained model parameters. We compared the Local Gradients UQ method to non-gradient measures defined using model probability outputs. The performance of each uncertainty measure was assessed in four clinically relevant experiments: (1) response to artificially degraded image quality, (2) comparison between matched high- and low-quality clinical images, (3) false positive (FP) filtering, and (4) correspondence with physician-rated disease likelihood.Main results.(1) Response to artificially degraded image quality was enhanced by the Local Gradients UQ method, where the median percent difference between matching lesions in non-degraded and most degraded images was consistently higher for the Local Gradients uncertainty measure than the non-gradient uncertainty measures (e.g. 62.35% vs. 2.16% for additive Gaussian noise). (2) The Local Gradients UQ measure responded better to high- and low-quality clinical images (p< 0.05 vsp> 0.1 for both non-gradient uncertainty measures). (3) FP filtering performance was enhanced by the Local Gradients UQ method when compared to the non-gradient methods, increasing the area under the receiver operating characteristic curve (ROC AUC) by 20.1% and decreasing the false positive rate by 26%. (4) The Local Gradients UQ method also showed more favorable correspondence with physician-rated likelihood for malignant lesions by increasing ROC AUC for correspondence with physician-rated disease likelihood by 16.2%.Significance. In summary, this work introduces and validates a novel gradient-based UQ method for deep learning-based medical image assessments to enhance user trust when using deployed clinical models.

Comparison of deep learning architectures for predicting amyloid positivity in Alzheimer's disease, mild cognitive impairment, and healthy aging, from T1-weighted brain structural MRI

Abnormal β-amyloid (Aβ) accumulation in the brain is an early indicator of Alzheimer's disease (AD) and is typically assessed through invasive procedures such as PET (positron emission tomography) or CSF (cerebrospinal fluid) assays. As new anti-Alzheimer's treatments can now successfully target amyloid pathology, there is a growing interest in predicting Aβ positivity (Aβ+) from less invasive, more widely available types of brain scans, such as T1-weighted (T1w) MRI. Here we compare multiple approaches to infer Aβ + from standard anatomical MRI: (1) classical machine learning algorithms, including logistic regression, XGBoost, and shallow artificial neural networks, (2) deep learning models based on 2D and 3D convolutional neural networks (CNNs), (3) a hybrid ANN-CNN, combining the strengths of shallow and deep neural networks, (4) transfer learning models based on CNNs, and (5) 3D Vision Transformers. All models were trained on paired MRI/PET data from 1,847 elderly participants (mean age: 75.1 yrs. ± 7.6SD; 863 females/984 males; 661 healthy controls, 889 with mild cognitive impairment (MCI), and 297 with Dementia), scanned as part of the Alzheimer's Disease Neuroimaging Initiative. We evaluated each model's balanced accuracy and F1 scores. While further tests on more diverse data are warranted, deep learning models trained on standard MRI showed promise for estimating Aβ + status, at least in people with MCI. This may offer a potential screening option before resorting to more invasive procedures.

DermSynth3D: Synthesis of in-the-wild annotated dermatology images

In recent years, deep learning (DL) has shown great potential in the field of dermatological image analysis. However, existing datasets in this domain have significant limitations, including a small number of image samples, limited disease conditions, insufficient annotations, and non-standardized image acquisitions. To address these shortcomings, we propose a novel framework called DermSynth3D. DermSynth3D blends skin disease patterns onto 3D textured meshes of human subjects using a differentiable renderer and generates 2D images from various camera viewpoints under chosen lighting conditions in diverse background scenes. Our method adheres to top-down rules that constrain the blending and rendering process to create 2D images with skin conditions that mimic in-the-wild acquisitions, ensuring more meaningful results. The framework generates photo-realistic 2D dermatological images and the corresponding dense annotations for semantic segmentation of the skin, skin conditions, body parts, bounding boxes around lesions, depth maps, and other 3D scene parameters, such as camera position and lighting conditions. DermSynth3D allows for the creation of custom datasets for various dermatology tasks. We demonstrate the effectiveness of data generated using DermSynth3D by training DL models on synthetic data and evaluating them on various dermatology tasks using real 2D dermatological images. We make our code publicly available at https://github.com/sfu-mial/DermSynth3D.

Evaluating Biases and Quality Issues in Intermodality Image Translation Studies for Neuroradiology: A Systematic Review

BACKGROUND

Intermodality image-to-image translation is an artificial intelligence technique for generating one technique from another.

PURPOSE

This review was designed to systematically identify and quantify biases and quality issues preventing validation and clinical application of artificial intelligence models for intermodality image-to-image translation of brain imaging.

DATA SOURCES

PubMed, Scopus, and IEEE Xplore were searched through August 2, 2023, for artificial intelligence-based image translation models of radiologic brain images.

STUDY SELECTION

This review collected 102 works published between April 2017 and August 2023.

DATA ANALYSIS

Eligible studies were evaluated for quality using the Checklist for Artificial Intelligence in Medical Imaging (CLAIM) and for bias using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). Medically-focused article adherence was compared with that of engineering-focused articles overall with the Mann-Whitney U test and for each criterion using the Fisher exact test.

DATA SYNTHESIS

Median adherence to the relevant CLAIM criteria was 69% and 38% for PROBAST questions. CLAIM adherence was lower for engineering-focused articles compared with medically-focused articles (65% versus 73%, P < .001). Engineering-focused studies had higher adherence for model description criteria, and medically-focused studies had higher adherence for data set and evaluation descriptions.

LIMITATIONS

Our review is limited by the study design and model heterogeneity.

CONCLUSIONS

Nearly all studies revealed critical issues preventing clinical application, with engineering-focused studies showing higher adherence for the technical model description but significantly lower overall adherence than medically-focused studies. The pursuit of clinical application requires collaboration from both fields to improve reporting.

Is deep learning-enabled real-time personalized CT dosimetry feasible using only patient images as input?

PURPOSE

To propose a novel deep-learning based dosimetry method that allows quick and accurate estimation of organ doses for individual patients, using only their computed tomography (CT) images as input.

METHODS

Despite recent advances in medical dosimetry, personalized CT dosimetry remains a labour-intensive process. Current state-of-the-art methods utilize time-consuming Monte Carlo (MC) based simulations for individual organ dose estimation in CT. The proposed method uses conditional generative adversarial networks (cGANs) to substitute MC simulations with fast dose image generation, based on image-to-image translation. The pix2pix architecture in conjunction with a regression model was utilized for the generation of the synthetic dose images. The lungs, heart, breast, bone and skin were manually segmented to estimate and compare organ doses calculated using both the original and synthetic dose images, respectively.

RESULTS

The average organ dose estimation error for the proposed method was 8.3% and did not exceed 20% for any of the organs considered. The performance of the method in the clinical environment was also assessed. Using segmentation tools developed in-house, an automatic organ dose calculation pipeline was set up. Calculation of organ doses for heart and lung for each CT slice took about 2 s.

CONCLUSIONS

This work shows that deep learning-enabled personalized CT dosimetry is feasible in real-time, using only patient CT images as input.

Personalized medicine in old age psychiatry and Alzheimer's disease

Elderly patients show us unfolded lives with unique individual characteristics. An increasing life span is associated with increasing physical and mental disease burden. Alzheimer's disease (AD) is an increasing challenge in old age. AD cannot be cured but it can be treated. The complexity of old age and AD offer targets for personalized medicine (PM). Targets for stratification of patients, detection of patients at risk for AD or for future targeted therapy are plentiful and can be found in several omic-levels.

Take CT, get PET free: AI-powered breakthrough in lung cancer diagnosis and prognosis

PET scans provide additional clinical value but are costly and not universally accessible. Salehjahromi et al.1 developed an AI-based pipeline to synthesize PET images from diagnostic CT scans, demonstrating its potential clinical utility across various clinical tasks for lung cancer.

A novel loss function to reproduce texture features for deep learning-based MRI-to-CT synthesis

BACKGROUND

Studies on computed tomography (CT) synthesis based on magnetic resonance imaging (MRI) have mainly focused on pixel-wise consistency, but the texture features of regions of interest (ROIs) have not received appropriate attention.

PURPOSE

This study aimed to propose a novel loss function to reproduce texture features of ROIs and pixel-wise consistency for deep learning-based MRI-to-CT synthesis. The method was expected to assist the multi-modality studies for radiomics.

METHODS

The study retrospectively enrolled 127 patients with nasopharyngeal carcinoma. CT and MRI images were collected for each patient, and then rigidly registered as pre-procession. We proposed a gray-level co-occurrence matrix (GLCM)-based loss function to improve the reproducibility of texture features. This novel loss function could be embedded into the present deep learning-based framework for image synthesis. In this study, a typical image synthesis model was selected as the baseline, which contained a Unet trained mean square error (MSE) loss function. We embedded the proposed loss function and designed experiments to supervise different ROIs to prove its effectiveness. The concordance correlation coefficient (CCC) of the GLCM feature was employed to evaluate the reproducibility of GLCM features, which are typical texture features. Besides, we used a publicly available dataset of brain tumors to verify our loss function.

RESULTS

Compared with the baseline, the proposed method improved the pixel-wise image quality metrics (MAE: 107.5 to 106.8 HU; SSIM: 0.9728 to 0.9730). CCC values of the GLCM features in GTVnx were significantly improved from 0.78 ± 0.12 to 0.82 ± 0.11 (p < 0.05 for paired t-test). Generally, > 90% (22/24) of the GLCM-based features were improved compared with the baseline, where the Informational Measure of Correlation feature was improved the most (CCC: 0.74 to 0.83). For the public dataset, the loss function also shows its effectiveness. With our proposed loss function added, the ability to reproduce texture features was improved in the ROIs.

CONCLUSIONS

The proposed method reproduced texture features for MRI-to-CT synthesis, which would benefit radiomics studies based on image multi-modality synthesis.

Clinical implementation of a commercial synthetic computed tomography solution for radiotherapy treatment of glioblastoma

Background and Purpose

Magnetic resonance (MR)-only radiotherapy (RT) workflow eliminates uncertainties due to computed tomography (CT)-MR image registration, by using synthetic CT (sCT) images generated from MR. This study describes the clinical implementation process, from retrospective commissioning to prospective validation stage of a commercial artificial intelligence (AI)-based sCT product. Evaluation of the dosimetric performance of the sCT is presented, with emphasis on the impact of voxel size differences between image modalities.

Materials and methods

sCT performance was assessed in glioblastoma RT planning. Dose differences for 30 patients in both commissioning and validation cohorts were calculated at various dose-volume-histogram (DVH) points for target and organs-at-risk (OAR). A gamma analysis was conducted on regridded image plans. Quality assurance (QA) guidelines were established based on commissioning phase results.

Results

Mean dose difference to target structures was found to be within ± 0.7 % regardless of image resolution and cohort. OARs' mean dose differences were within ± 1.3 % for plans calculated on regridded images for both cohorts, while differences were higher for plans with original voxel size, reaching up to -4.2 % for chiasma D2% in the commissioning cohort. Gamma passing rates for the brain structure using the criteria 1 %/1mm, 2 %/2mm and 3 %/3mm were 93.6 %/99.8 %/100 % and 96.6 %/99.9 %/100 % for commissioning and validation cohorts, respectively.

Conclusions

Dosimetric outcomes in both commissioning and validation stages confirmed sCT's equivalence to CT. The large patient cohort in this study aided in establishing a robust QA program for the MR-only workflow, now applied in glioblastoma RT at our center.

A systematic literature review: deep learning techniques for synthetic medical image generation and their applications in radiotherapy

The aim of this systematic review is to determine whether Deep Learning (DL) algorithms can provide a clinically feasible alternative to classic algorithms for synthetic Computer Tomography (sCT). The following categories are presented in this study: ∙ MR-based treatment planning and synthetic CT generation techniques. ∙ Generation of synthetic CT images based on Cone Beam CT images. ∙ Low-dose CT to High-dose CT generation. ∙ Attenuation correction for PET images. To perform appropriate database searches, we reviewed journal articles published between January 2018 and June 2023. Current methodology, study strategies, and results with relevant clinical applications were analyzed as we outlined the state-of-the-art of deep learning based approaches to inter-modality and intra-modality image synthesis. This was accomplished by contrasting the provided methodologies with traditional research approaches. The key contributions of each category were highlighted, specific challenges were identified, and accomplishments were summarized. As a final step, the statistics of all the cited works from various aspects were analyzed, which revealed that DL-based sCTs have achieved considerable popularity, while also showing the potential of this technology. In order to assess the clinical readiness of the presented methods, we examined the current status of DL-based sCT generation.

Synthetic PET from CT improves diagnosis and prognosis for lung cancer: Proof of concept

[18F]Fluorodeoxyglucose positron emission tomography (FDG-PET) and computed tomography (CT) are indispensable components in modern medicine. Although PET can provide additional diagnostic value, it is costly and not universally accessible, particularly in low-income countries. To bridge this gap, we have developed a conditional generative adversarial network pipeline that can produce FDG-PET from diagnostic CT scans based on multi-center multi-modal lung cancer datasets (n = 1,478). Synthetic PET images are validated across imaging, biological, and clinical aspects. Radiologists confirm comparable imaging quality and tumor contrast between synthetic and actual PET scans. Radiogenomics analysis further proves that the dysregulated cancer hallmark pathways of synthetic PET are consistent with actual PET. We also demonstrate the clinical values of synthetic PET in improving lung cancer diagnosis, staging, risk prediction, and prognosis. Taken together, this proof-of-concept study testifies to the feasibility of applying deep learning to obtain high-fidelity PET translated from CT.

Simulation of acquisition shifts in T2 weighted fluid-attenuated inversion recovery magnetic resonance images to stress test artificial intelligence segmentation networks

Purpose

To provide a simulation framework for routine neuroimaging test data, which allows for "stress testing" of deep segmentation networks against acquisition shifts that commonly occur in clinical practice for T2 weighted (T2w) fluid-attenuated inversion recovery magnetic resonance imaging protocols.

Approach

The approach simulates "acquisition shift derivatives" of MR images based on MR signal equations. Experiments comprise the validation of the simulated images by real MR scans and example stress tests on state-of-the-art multiple sclerosis lesion segmentation networks to explore a generic model function to describe the F1 score in dependence of the contrast-affecting sequence parameters echo time (TE) and inversion time (TI).

Results

The differences between real and simulated images range up to 19% in gray and white matter for extreme parameter settings. For the segmentation networks under test, the F1 score dependency on TE and TI can be well described by quadratic model functions (R 2 > 0.9 ). The coefficients of the model functions indicate that changes of TE have more influence on the model performance than TI.

Conclusions

We show that these deviations are in the range of values as may be caused by erroneous or individual differences in relaxation times as described by literature. The coefficients of the F1 model function allow for a quantitative comparison of the influences of TE and TI. Limitations arise mainly from tissues with a low baseline signal (like cerebrospinal fluid) and when the protocol contains contrast-affecting measures that cannot be modeled due to missing information in the DICOM header.

Segmentation of pelvic structures in T2 MRI via MR-to-CT synthesis

Segmentation of multiple pelvic structures in MRI volumes is a prerequisite for many clinical applications, such as sarcopenia assessment, bone density measurement, and muscle-to-fat volume ratio estimation. While many CT-specific datasets and automated CT-based multi-structure pelvis segmentation methods exist, there are few MRI-specific multi-structure segmentation methods in literature. In this pilot work, we propose a lightweight and annotation-free pipeline to synthetically translate T2 MRI volumes of the pelvis to CT, and subsequently leverage an existing CT-only tool called TotalSegmentator to segment 8 pelvic structures in the generated CT volumes. The predicted masks were then mapped back to the original MR volumes as segmentation masks. We compared the predicted masks against the expert annotations of the public TCGA-UCEC dataset and an internal dataset. Experiments demonstrated that the proposed pipeline achieved Dice measures ≥65% for 8 pelvic structures in T2 MRI. The proposed pipeline is an alternative method to obtain multi-organ and structure segmentations without being encumbered by time-consuming manual annotations. By exploiting the significant research progress in CTs, it is possible to extend the proposed pipeline to other MRI sequences in principle. Our research bridges the chasm between the current CT-based multi-structure segmentation and MRI-based segmentation. The manually segmented structures in the TCGA-UCEC dataset are publicly available.

Attenuation correction and truncation completion for breast PET/MR imaging using deep learning

Objective. Simultaneous PET/MR scanners combine the high sensitivity of MR imaging with the functional imaging of PET. However, attenuation correction of breast PET/MR imaging is technically challenging. The purpose of this study is to establish a robust attenuation correction algorithm for breast PET/MR images that relies on deep learning (DL) to recreate the missing portions of the patient's anatomy (truncation completion), as well as to provide bone information for attenuation correction from only the PET data.Approach. Data acquired from 23 female subjects with invasive breast cancer scanned with18F-fluorodeoxyglucose PET/CT and PET/MR localized to the breast region were used for this study. Three DL models, U-Net with mean absolute error loss (DLMAE) model, U-Net with mean squared error loss (DLMSE) model, and U-Net with perceptual loss (DLPerceptual) model, were trained to predict synthetic CT images (sCT) for PET attenuation correction (AC) given non-attenuation corrected (NAC) PETPET/MRimages as inputs. The DL and Dixon-based sCT reconstructed PET images were compared against those reconstructed from CT images by calculating the percent error of the standardized uptake value (SUV) and conducting Wilcoxon signed rank statistical tests.Main results. sCT images from the DLMAEmodel, the DLMSEmodel, and the DLPerceptualmodel were similar in mean absolute error (MAE), peak-signal-to-noise ratio, and normalized cross-correlation. No significant difference in SUV was found between the PET images reconstructed using the DLMSEand DLPerceptualsCTs compared to the reference CT for AC in all tissue regions. All DL methods performed better than the Dixon-based method according to SUV analysis.Significance. A 3D U-Net with MSE or perceptual loss model can be implemented into a reconstruction workflow, and the derived sCT images allow successful truncation completion and attenuation correction for breast PET/MR images.

Detection support of lesions in patients with prostate cancer using [Formula: see text]-PSMA 1007 PET/CT

PURPOSE

This study proposes a detection support system for primary and metastatic lesions of prostate cancer using [Formula: see text]-PSMA 1007 positron emission tomography/computed tomography (PET/CT) images with non-image information, including patient metadata and location information of an input slice image.

METHODS

A convolutional neural network with condition generators and feature-wise linear modulation (FiLM) layers was employed to allow input of not only PET/CT images but also non-image information, namely, Gleason score, flag of pre- or post-prostatectomy, and normalized z-coordinate of an input slice. We explored the insertion position of the FiLM layers to optimize the conditioning of the network using non-image information.

RESULTS

[Formula: see text]-PSMA 1007 PET/CT images were collected from 163 patients with prostate cancer and applied to the proposed system in a threefold cross-validation manner to evaluate the performance. The proposed system achieved a Dice score of 0.5732 (per case) and sensitivity of 0.8200 (per lesion), which are 3.87 and 4.16 points higher than the network without non-image information.

CONCLUSION

This study demonstrated the effectiveness of the use of non-image information, including metadata of the patient and location information of the input slice image, in the detection of prostate cancer from [Formula: see text]-PSMA 1007 PET/CT images. Improvement in the sensitivity of inactive and small lesions remains a future challenge.

Deep learning based synthesis of MRI, CT and PET: Review and analysis

Medical image synthesis represents a critical area of research in clinical decision-making, aiming to overcome the challenges associated with acquiring multiple image modalities for an accurate clinical workflow. This approach proves beneficial in estimating an image of a desired modality from a given source modality among the most common medical imaging contrasts, such as Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and Positron Emission Tomography (PET). However, translating between two image modalities presents difficulties due to the complex and non-linear domain mappings. Deep learning-based generative modelling has exhibited superior performance in synthetic image contrast applications compared to conventional image synthesis methods. This survey comprehensively reviews deep learning-based medical imaging translation from 2018 to 2023 on pseudo-CT, synthetic MR, and synthetic PET. We provide an overview of synthetic contrasts in medical imaging and the most frequently employed deep learning networks for medical image synthesis. Additionally, we conduct a detailed analysis of each synthesis method, focusing on their diverse model designs based on input domains and network architectures. We also analyse novel network architectures, ranging from conventional CNNs to the recent Transformer and Diffusion models. This analysis includes comparing loss functions, available datasets and anatomical regions, and image quality assessments and performance in other downstream tasks. Finally, we discuss the challenges and identify solutions within the literature, suggesting possible future directions. We hope that the insights offered in this survey paper will serve as a valuable roadmap for researchers in the field of medical image synthesis.

CT-based generation of synthetic-pseudo MR images with different weightings for human knee

Synthetic MR images are generated for their high soft-tissue contrast avoiding the discomfort by the long acquisition time and placing claustrophobic patients in the MR scanner's confined space. The aim of this study is to generate synthetic pseudo-MR images from a real CT image for the knee region in vivo. 19 healthy subjects were scanned for model training, while 13 other healthy subjects were imaged for testing. The approach used in this work is novel such that the registration was performed between the MR and CT images, and the femur bone, patella, and the surrounding soft tissue were segmented on the CT image. The tissue type was mapped to its corresponding mean and standard deviation values of the CT# of a window moving on each pixel in the reconstructed CT images, which enabled the remapping of the tissue to its MRI intrinsic parameters: T1, T2, and proton density (ρ). To generate the synthetic MR image of a knee slice, a classic spin-echo sequence was simulated using proper intrinsic and contrast parameters. Results showed that the synthetic MR images were comparable to the real images acquired with the same TE and TR values, and the average slope between them (for all knee segments) was 0.98, while the average percentage root mean square difference (PRD) was 25.7%. In conclusion, this study has shown the feasibility and validity of accurately generating synthetic MR images of the knee region in vivo with different weightings from a single real CT image.

Application of geometric shape-based CT field-of-view extension algorithms in an all-digital positron emission tomography/computed tomography system

BACKGROUND

Computed tomography (CT)-based positron emission tomography (PET) attenuation correction (AC) is a commonly used method in PET AC. However, the CT truncation caused by the subject's limbs outside the CT field-of-view (FOV) leads to errors in PET AC.

PURPOSE

In order to enhance the quantitative accuracy of PET imaging in the all-digital DigitMI 930 PET/CT system, we assessed the impact of FOV truncation on its image quality and investigated the effectiveness of geometric shape-based FOV extension algorithms in this system.

METHODS

We implemented two geometric shape-based FOV extension algorithms. By setting the data from different numbers of detector channels on either side of the sinogram to zero, we simulated various levels of truncation. Specific regions of interest (ROI) were selected, and the mean values of these ROIs were calculated to visually compare the differences between truncated CT, CT extended using the FOV extension algorithms, and the original CT. Furthermore, we conducted statistical analyses on the mean and standard deviation of residual maps between truncated/extended CT and the original CT at different levels of truncation. Subsequently, similar data processing was applied to PET images corrected using original CT and those corrected using simulated truncated and extended CT images. This allowed us to evaluate the influence of FOV truncation on the images produced by the DigitMI 930 PET/CT system and assess the effectiveness of the FOV extension algorithms.

RESULTS

Truncation caused bright artifacts at the CT FOV edge and a slight increase in pixel values within the FOV. When using truncated CT data for PET AC, the PET activity outside the CT FOV decreased, while the extension algorithm effectively reduced these effects. Patient data showed that the activity within the CT FOV decreased by 60% in the truncated image compared to the base image, but this number could be reduced to at least 17.3% after extension.

CONCLUSION

The two geometric shape-based algorithms effectively eliminate CT truncation artifacts and restore the true distribution of CT shape and PET emission data outside the FOV in the all-digital DigitMI 930 PET/CT system. These two algorithms can be used as basic solutions for CT FOV extension in all-digital PET/CT systems.

Deep learning in MRI-guided radiation therapy: A systematic review

Recent advances in MRI-guided radiation therapy (MRgRT) and deep learning techniques encourage fully adaptive radiation therapy (ART), real-time MRI monitoring, and the MRI-only treatment planning workflow. Given the rapid growth and emergence of new state-of-the-art methods in these fields, we systematically review 197 studies written on or before December 31, 2022, and categorize the studies into the areas of image segmentation, image synthesis, radiomics, and real time MRI. Building from the underlying deep learning methods, we discuss their clinical importance and current challenges in facilitating small tumor segmentation, accurate x-ray attenuation information from MRI, tumor characterization and prognosis, and tumor motion tracking. In particular, we highlight the recent trends in deep learning such as the emergence of multi-modal, visual transformer, and diffusion models.

Artificial Intelligence's Transformative Role in Illuminating Brain Function in Long COVID Patients Using PET/FDG

Cutting-edge brain imaging techniques, particularly positron emission tomography with Fluorodeoxyglucose (PET/FDG), are being used in conjunction with Artificial Intelligence (AI) to shed light on the neurological symptoms associated with Long COVID. AI, particularly deep learning algorithms such as convolutional neural networks (CNN) and generative adversarial networks (GAN), plays a transformative role in analyzing PET scans, identifying subtle metabolic changes, and offering a more comprehensive understanding of Long COVID's impact on the brain. It aids in early detection of abnormal brain metabolism patterns, enabling personalized treatment plans. Moreover, AI assists in predicting the progression of neurological symptoms, refining patient care, and accelerating Long COVID research. It can uncover new insights, identify biomarkers, and streamline drug discovery. Additionally, the application of AI extends to non-invasive brain stimulation techniques, such as transcranial direct current stimulation (tDCS), which have shown promise in alleviating Long COVID symptoms. AI can optimize treatment protocols by analyzing neuroimaging data, predicting individual responses, and automating adjustments in real time. While the potential benefits are vast, ethical considerations and data privacy must be rigorously addressed. The synergy of AI and PET scans in Long COVID research offers hope in understanding and mitigating the complexities of this condition.

Uncertainty quantification in computed tomography pulmonary angiography

Computed tomography (CT) imaging of the thorax is widely used for the detection and monitoring of pulmonary embolism (PE). However, CT images can contain artifacts due to the acquisition or the processes involved in image reconstruction. Radiologists often have to distinguish between such artifacts and actual PEs. We provide a proof of concept in the form of a scalable hypothesis testing method for CT, to enable quantifying uncertainty of possible PEs. In particular, we introduce a Bayesian Framework to quantify the uncertainty of an observed compact structure that can be identified as a PE. We assess the ability of the method to operate under high-noise environments and with insufficient data.

Multi-contrast high-field quality image synthesis for portable low-field MRI using generative adversarial networks and paired data

Introduction

Portable low-field strength (64mT) MRI scanners promise to increase access to neuroimaging for clinical and research purposes, however these devices produce lower quality images compared to high-field scanners. In this study, we developed and evaluated a deep learning architecture to generate high-field quality brain images from low-field inputs using a paired dataset of multiple sclerosis (MS) patients scanned at 64mT and 3T.

Methods

A total of 49 MS patients were scanned on portable 64mT and standard 3T scanners at Penn (n=25) or the National Institutes of Health (NIH, n=24) with T1-weighted, T2-weighted and FLAIR acquisitions. Using this paired data, we developed a generative adversarial network (GAN) architecture for low- to high-field image translation (LowGAN). We then evaluated synthesized images with respect to image quality, brain morphometry, and white matter lesions.

Results

Synthetic high-field images demonstrated visually superior quality compared to low-field inputs and significantly higher normalized cross-correlation (NCC) to actual high-field images for T1 (p=0.001) and FLAIR (p<0.001) contrasts. LowGAN generally outperformed the current state-of-the-art for low-field volumetrics. For example, thalamic, lateral ventricle, and total cortical volumes in LowGAN outputs did not differ significantly from 3T measurements. Synthetic outputs preserved MS lesions and captured a known inverse relationship between total lesion volume and thalamic volume.

Conclusions

LowGAN generates synthetic high-field images with comparable visual and quantitative quality to actual high-field scans. Enhancing portable MRI image quality could add value and boost clinician confidence, enabling wider adoption of this technology.

Synthetic Knee MRI T1p Maps as an Avenue for Clinical Translation of Quantitative Osteoarthritis Biomarkers

A 2D U-Net was trained to generate synthetic T1p maps from T2 maps for knee MRI to explore the feasibility of domain adaptation for enriching existing datasets and enabling rapid, reliable image reconstruction. The network was developed using 509 healthy contralateral and injured ipsilateral knee images from patients with ACL injuries and reconstruction surgeries acquired across three institutions. Network generalizability was evaluated on 343 knees acquired in a clinical setting and 46 knees from simultaneous bilateral acquisition in a research setting. The deep neural network synthesized high-fidelity reconstructions of T1p maps, preserving textures and local T1p elevation patterns in cartilage with a normalized mean square error of 2.4% and Pearson's correlation coefficient of 0.93. Analysis of reconstructed T1p maps within cartilage compartments revealed minimal bias (-0.10 ms), tight limits of agreement, and quantification error (5.7%) below the threshold for clinically significant change (6.42%) associated with osteoarthritis. In an out-of-distribution external test set, synthetic maps preserved T1p textures, but exhibited increased bias and wider limits of agreement. This study demonstrates the capability of image synthesis to reduce acquisition time, derive meaningful information from existing datasets, and suggest a pathway for standardizing T1p as a quantitative biomarker for osteoarthritis.

How Artificial Intelligence Is Shaping Medical Imaging Technology: A Survey of Innovations and Applications

The integration of artificial intelligence (AI) into medical imaging has guided in an era of transformation in healthcare. This literature review explores the latest innovations and applications of AI in the field, highlighting its profound impact on medical diagnosis and patient care. The innovation segment explores cutting-edge developments in AI, such as deep learning algorithms, convolutional neural networks, and generative adversarial networks, which have significantly improved the accuracy and efficiency of medical image analysis. These innovations have enabled rapid and accurate detection of abnormalities, from identifying tumors during radiological examinations to detecting early signs of eye disease in retinal images. The article also highlights various applications of AI in medical imaging, including radiology, pathology, cardiology, and more. AI-based diagnostic tools not only speed up the interpretation of complex images but also improve early detection of disease, ultimately delivering better outcomes for patients. Additionally, AI-based image processing facilitates personalized treatment plans, thereby optimizing healthcare delivery. This literature review highlights the paradigm shift that AI has brought to medical imaging, highlighting its role in revolutionizing diagnosis and patient care. By combining cutting-edge AI techniques and their practical applications, it is clear that AI will continue shaping the future of healthcare in profound and positive ways.

A Critical Assessment of Generative Models for Synthetic Data Augmentation on Limited Pneumonia X-ray Data

In medical imaging, deep learning models serve as invaluable tools for expediting diagnoses and aiding specialized medical professionals in making clinical decisions. However, effectively training deep learning models typically necessitates substantial quantities of high-quality data, a resource often lacking in numerous medical imaging scenarios. One way to overcome this deficiency is to artificially generate such images. Therefore, in this comparative study we train five generative models to artificially increase the amount of available data in such a scenario. This synthetic data approach is evaluated on a a downstream classification task, predicting four causes for pneumonia as well as healthy cases on 1082 chest X-ray images. Quantitative and medical assessments show that a Generative Adversarial Network (GAN)-based approach significantly outperforms more recent diffusion-based approaches on this limited dataset with better image quality and pathological plausibility. We show that better image quality surprisingly does not translate to improved classification performance by evaluating five different classification models and varying the amount of additional training data. Class-specific metrics like precision, recall, and F1-score show a substantial improvement by using synthetic images, emphasizing the data rebalancing effect of less frequent classes. However, overall performance does not improve for most models and configurations, except for a DreamBooth approach which shows a +0.52 improvement in overall accuracy. The large variance of performance impact in this study suggests a careful consideration of utilizing generative models for limited data scenarios, especially with an unexpected negative correlation between image quality and downstream classification improvement.

Prior information-guided reconstruction network for positron emission tomography images

Background

Deep learning has recently shown great potential in medical image reconstruction tasks. For positron emission tomography (PET) images, the direct reconstruction from raw data to radioactivity images using deep learning without any constraint may lead to the production of nonexistent structures. The aim of this study was to specifically develop and test a flexibly deep learning-based reconstruction network guided by any form of prior knowledge to achieve high quality and high reliability reconstruction.

Methods

We developed a novel prior information-guided reconstruction network (PIGRN) with a dual-channel generator and a 2-scale discriminator based on a conditional generative adversarial network (cGAN). Besides the raw data channel, an additional channel is provided in the generator for prior information (PI) to guide the training phase. The PI can be reconstructed images obtained via conventional methods, nuclear medical images from other modalities, attenuation correction maps from time-of-flight-PET (TOF-PET) data, or any other physical parameters. For this study, the reconstructed images generated by filtered back projection (FBP) were chosen as the input of the additional channel. To improve the image quality, a 2-scale discriminator was adopted which can focus on both the coarse and fine field of the reconstruction images. Experiments were carried out on both a simulation dataset and a real Sprague Dawley (SD) rat dataset.

Results

Two classic deep learning-based reconstruction networks, including U-Net and Deep-PET, were compared in our study. Compared with these two methods, our method could provide much higher quality PET image reconstruction in the study of the simulation dataset. The peak signal-to-noise ratio (PSNR) value reached 31.8498, and the structure similarity index measure (SSIM) value reached 0.9754. The real study on SD rats indicated that the proposed network also has strong generalization ability.

Conclusions

The flexible PIGRN based on cGAN for PET images combines both raw data and PI. The results of comparison experiments and a generalization experiment based on simulation and SD rat datasets demonstrated that the proposed PIGRN has the ability to improve image quality and has strong generalization ability.

Deformation equivariant cross-modality image synthesis with paired non-aligned training data

Cross-modality image synthesis is an active research topic with multiple medical clinically relevant applications. Recently, methods allowing training with paired but misaligned data have started to emerge. However, no robust and well-performing methods applicable to a wide range of real world data sets exist. In this work, we propose a generic solution to the problem of cross-modality image synthesis with paired but non-aligned data by introducing new deformation equivariance encouraging loss functions. The method consists of joint training of an image synthesis network together with separate registration networks and allows adversarial training conditioned on the input even with misaligned data. The work lowers the bar for new clinical applications by allowing effortless training of cross-modality image synthesis networks for more difficult data sets.

Denoising diffusion-based MRI to CT image translation enables automated spinal segmentation

BACKGROUND

Automated segmentation of spinal magnetic resonance imaging (MRI) plays a vital role both scientifically and clinically. However, accurately delineating posterior spine structures is challenging.

METHODS

This retrospective study, approved by the ethical committee, involved translating T1-weighted and T2-weighted images into computed tomography (CT) images in a total of 263 pairs of CT/MR series. Landmark-based registration was performed to align image pairs. We compared two-dimensional (2D) paired - Pix2Pix, denoising diffusion implicit models (DDIM) image mode, DDIM noise mode - and unpaired (SynDiff, contrastive unpaired translation) image-to-image translation using "peak signal-to-noise ratio" as quality measure. A publicly available segmentation network segmented the synthesized CT datasets, and Dice similarity coefficients (DSC) were evaluated on in-house test sets and the "MRSpineSeg Challenge" volumes. The 2D findings were extended to three-dimensional (3D) Pix2Pix and DDIM.

RESULTS

2D paired methods and SynDiff exhibited similar translation performance and DCS on paired data. DDIM image mode achieved the highest image quality. SynDiff, Pix2Pix, and DDIM image mode demonstrated similar DSC (0.77). For craniocaudal axis rotations, at least two landmarks per vertebra were required for registration. The 3D translation outperformed the 2D approach, resulting in improved DSC (0.80) and anatomically accurate segmentations with higher spatial resolution than that of the original MRI series.

CONCLUSIONS

Two landmarks per vertebra registration enabled paired image-to-image translation from MRI to CT and outperformed all unpaired approaches. The 3D techniques provided anatomically correct segmentations, avoiding underprediction of small structures like the spinous process.

RELEVANCE STATEMENT

This study addresses the unresolved issue of translating spinal MRI to CT, making CT-based tools usable for MRI data. It generates whole spine segmentation, previously unavailable in MRI, a prerequisite for biomechanical modeling and feature extraction for clinical applications.

KEY POINTS

• Unpaired image translation lacks in converting spine MRI to CT effectively. • Paired translation needs registration with two landmarks per vertebra at least. • Paired image-to-image enables segmentation transfer to other domains. • 3D translation enables super resolution from MRI to CT. • 3D translation prevents underprediction of small structures.

Comprehensive evaluation of similarity between synthetic and real CT images for nasopharyngeal carcinoma

BACKGROUND

Although magnetic resonance imaging (MRI)-to-computed tomography (CT) synthesis studies based on deep learning have significantly progressed, the similarity between synthetic CT (sCT) and real CT (rCT) has only been evaluated in image quality metrics (IQMs). To evaluate the similarity between synthetic CT (sCT) and real CT (rCT) comprehensively, we comprehensively evaluated IQMs and radiomic features for the first time.

METHODS

This study enrolled 127 patients with nasopharyngeal carcinoma who underwent CT and MRI scans. Supervised-learning (Unet) and unsupervised-learning (CycleGAN) methods were applied to build MRI-to-CT synthesis models. The regions of interest (ROIs) included nasopharynx gross tumor volume (GTVnx), brainstem, parotid glands, and temporal lobes. The peak signal-to-noise ratio (PSNR), mean absolute error (MAE), root mean square error (RMSE), and structural similarity (SSIM) were used to evaluate image quality. Additionally, 837 radiomic features were extracted for each ROI, and the correlation was evaluated using the concordance correlation coefficient (CCC).

RESULTS

The MAE, RMSE, SSIM, and PSNR of the body were 91.99, 187.12, 0.97, and 51.15 for Unet and 108.30, 211.63, 0.96, and 49.84 for CycleGAN. For the metrics, Unet was superior to CycleGAN (P < 0.05). For the radiomic features, the percentage of four levels (i.e., excellent, good, moderate, and poor, respectively) were as follows: GTVnx, 8.5%, 14.6%, 26.5%, and 50.4% for Unet and 12.3%, 25%, 38.4%, and 24.4% for CycleGAN; other ROIs, 5.44% ± 3.27%, 5.56% ± 2.92%, 21.38% ± 6.91%, and 67.58% ± 8.96% for Unet and 5.16% ± 1.69%, 3.5% ± 1.52%, 12.68% ± 7.51%, and 78.62% ± 8.57% for CycleGAN.

CONCLUSIONS

Unet-sCT was superior to CycleGAN-sCT for the IQMs. However, neither exhibited absolute superiority in radiomic features, and both were far less similar to rCT. Therefore, further work is required to improve the radiomic similarity for MRI-to-CT synthesis.

TRIAL REGISTRATION

This study was a retrospective study, so it was free from registration.

Efficient segmentation using domain adaptation for MRI-guided and CBCT-guided online adaptive radiotherapy

BACKGROUND

Delineation of regions of interest (ROIs) is important for adaptive radiotherapy (ART) but it is also time consuming and labor intensive.

AIM

This study aims to develop efficient segmentation methods for magnetic resonance imaging-guided ART (MRIgART) and cone-beam computed tomography-guided ART (CBCTgART).

MATERIALS AND METHODS

MRIgART and CBCTgART studies enrolled 242 prostate cancer patients and 530 nasopharyngeal carcinoma patients, respectively. A public dataset of CBCT from 35 pancreatic cancer patients was adopted to test the framework. We designed two domain adaption methods to learn and adapt the features from planning computed tomography (pCT) to MRI or CBCT modalities. The pCT was transformed to synthetic MRI (sMRI) for MRIgART, while CBCT was transformed to synthetic CT (sCT) for CBCTgART. Generalized segmentation models were trained with large popular data in which the inputs were sMRI for MRIgART and pCT for CBCTgART. Finally, the personalized models for each patient were established by fine-tuning the generalized model with the contours on pCT of that patient. The proposed method was compared with deformable image registration (DIR), a regular deep learning (DL) model trained on the same modality (DL-regular), and a generalized model in our framework (DL-generalized).

RESULTS

The proposed method achieved better or comparable performance. For MRIgART of the prostate cancer patients, the mean dice similarity coefficient (DSC) of four ROIs was 87.2%, 83.75%, 85.36%, and 92.20% for the DIR, DL-regular, DL-generalized, and proposed method, respectively. For CBCTgART of the nasopharyngeal carcinoma patients, the mean DSC of two target volumes were 90.81% and 91.18%, 75.17% and 58.30%, for the DIR, DL-regular, DL-generalized, and the proposed method, respectively. For CBCTgART of the pancreatic cancer patients, the mean DSC of two ROIs were 61.94% and 61.44%, 63.94% and 81.56%, for the DIR, DL-regular, DL-generalized, and the proposed method, respectively.

CONCLUSION

The proposed method utilizing personalized modeling improved the segmentation accuracy of ART.

Bone-GAN: Generation of virtual bone microstructure of high resolution peripheral quantitative computed tomography

BACKGROUND

Data-driven development of medical biomarkers of bone requires a large amount of image data but physical measurements are generally too restricted in size and quality to perform a robust training.

PURPOSE

This study aims to provide a reliable in silico method for the generation of realistic bone microstructure with defined microarchitectural properties. Synthetic bone samples may improve training of neural networks and serve for the development of new diagnostic parameters of bone architecture and mineralization.

METHODS

One hundred-fifty cadaveric lumbar vertebrae from 48 different male human spines were scanned with a high resolution peripheral quantitative CT. After prepocessing the scans, we extracted 10,795 purely spongeous bone patches, each with a side length of 32 voxels (5 mm) and isotropic voxel size of 164 μm. We trained a volumetric generative adversarial network (GAN) in a progressive manner to create synthetic microstructural bone samples. We then added a style transfer technique to allow the generation of synthetic samples with defined microstructure and gestalt by simultaneously optimizing two entangled loss functions. Reliability testing was performed by comparing real and synthetic bone samples on 10 well-understood microstructural parameters.

RESULTS

The method was able to create synthetic bone samples with visual and quantitative properties that effectively matched with the real samples. The GAN contained a well-formed latent space allowing to smoothly morph bone samples by their microstructural parameters, visual appearance or both. Optimum performance has been obtained for bone samples with voxel size 32 × 32 × 32, but also samples of size 64 × 64 × 64 could be synthesized.

CONCLUSIONS

Our two-step-approach combines a parameter-agnostic GAN with a parameter-specific style transfer technique. It allows to generate an unlimited anonymous database of microstructural bone samples with sufficient realism to be used for the development of new data-driven methods of bone-biomarkers. Particularly, the style transfer technique can generate datasets of bone samples with specific conditions to simulate certain bone pathologies.

Unified Brain MR-Ultrasound Synthesis using Multi-Modal Hierarchical Representations

We introduce MHVAE, a deep hierarchical variational autoencoder (VAE) that synthesizes missing images from various modalities. Extending multi-modal VAEs with a hierarchical latent structure, we introduce a probabilistic formulation for fusing multi-modal images in a common latent representation while having the flexibility to handle incomplete image sets as input. Moreover, adversarial learning is employed to generate sharper images. Extensive experiments are performed on the challenging problem of joint intra-operative ultrasound (iUS) and Magnetic Resonance (MR) synthesis. Our model outperformed multi-modal VAEs, conditional GANs, and the current state-of-the-art unified method (ResViT) for synthesizing missing images, demonstrating the advantage of using a hierarchical latent representation and a principled probabilistic fusion operation. Our code is publicly available.

Retrospective T2 quantification from conventional weighted MRI of the prostate based on deep learning

Purpose

To develop a deep learning-based method to retrospectively quantify T2 from conventional T1- and T2-weighted images.

Methods

Twenty-five subjects were imaged using a multi-echo spin-echo sequence to estimate reference prostate T2 maps. Conventional T1- and T2-weighted images were acquired as the input images. A U-Net based neural network was developed to directly estimate T2 maps from the weighted images using a four-fold cross-validation training strategy. The structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), mean percentage error (MPE), and Pearson correlation coefficient were calculated to evaluate the quality of network-estimated T2 maps. To explore the potential of this approach in clinical practice, a retrospective T2 quantification was performed on a high-risk prostate cancer cohort (Group 1) and a low-risk active surveillance cohort (Group 2). Tumor and non-tumor T2 values were evaluated by an experienced radiologist based on region of interest (ROI) analysis.

Results

The T2 maps generated by the trained network were consistent with the corresponding reference. Prostate tissue structures and contrast were well preserved, with a PSNR of 26.41 ± 1.17 dB, an SSIM of 0.85 ± 0.02, and a Pearson correlation coefficient of 0.86. Quantitative ROI analyses performed on 38 prostate cancer patients revealed estimated T2 values of 80.4 ± 14.4 ms and 106.8 ± 16.3 ms for tumor and non-tumor regions, respectively. ROI measurements showed a significant difference between tumor and non-tumor regions of the estimated T2 maps (P < 0.001). In the two-timepoints active surveillance cohort, patients defined as progressors exhibited lower estimated T2 values of the tumor ROIs at the second time point compared to the first time point. Additionally, the T2 difference between two time points for progressors was significantly greater than that for non-progressors (P = 0.010).

Conclusion

A deep learning method was developed to estimate prostate T2 maps retrospectively from clinically acquired T1- and T2-weighted images, which has the potential to improve prostate cancer diagnosis and characterization without requiring extra scans.

Generation of synthetic CT from CBCT using deep learning approaches for head and neck cancer patients

Purpose.To create a synthetic CT (sCT) from daily CBCT using either deep residual U-Net (DRUnet), or conditional generative adversarial network (cGAN) for adaptive radiotherapy planning (ART).Methods.First fraction CBCT and planning CT (pCT) were collected from 93 Head and Neck patients who underwent external beam radiotherapy. The dataset was divided into training, validation, and test sets of 58, 10 and 25 patients respectively. Three methods were used to generate sCT, 1. Nonlocal means patch based method was modified to include multiscale patches defining the multiscale patch based method (MPBM), 2. An encoder decoder 2D Unet with imbricated deep residual units was implemented, 3. DRUnet was integrated to the generator part of cGAN whereas a convolutional PatchGAN classifier was used as the discriminator. The accuracy of sCT was evaluated geometrically using Mean Absolute Error (MAE). Clinical Volumetric Modulated Arc Therapy (VMAT) plans were copied from pCT to registered CBCT and sCT and dosimetric analysis was performed by comparing Dose Volume Histogram (DVH) parameters of planning target volumes (PTVs) and organs at risk (OARs). Furthermore, 3D Gamma analysis (2%/2mm, global) between the dose on the sCT or CBCT and that on the pCT was performed.Results. The average MAE calculated between pCT and CBCT was 180.82 ± 27.37HU. Overall, all approaches significantly reduced the uncertainties in CBCT. Deep learning approaches outperformed patch-based methods with MAE = 67.88 ± 8.39HU (DRUnet) and MAE = 72.52 ± 8.43HU (cGAN) compared to MAE = 90.69 ± 14.3HU (MPBM). The percentages of DVH metric deviations were below 0.55% for PTVs and 1.17% for OARs using DRUnet. The average Gamma pass rate was 99.45 ± 1.86% for sCT generated using DRUnet.Conclusion.DL approaches outperformed MPBM. Specifically, DRUnet could be used for the generation of sCT with accurate intensities and realistic description of patient anatomy. This could be beneficial for CBCT based ART.

Deep learning based brain MRI registration driven by local-signed-distance fields of segmentation maps

BACKGROUND

Deep learning based unsupervised registration utilizes the intensity information to align images. To avoid the influence of intensity variation and improve the registration accuracy, unsupervised and weakly-supervised registration are combined, namely, dually-supervised registration. However, the estimated dense deformation fields (DDFs) will focus on the edges among adjacent tissues when the segmentation labels are directly used to drive the registration progress, which will decrease the plausibility of brain MRI registration.

PURPOSE

In order to increase the accuracy of registration and ensure the plausibility of registration at the same time, we combine the local-signed-distance fields (LSDFs) and intensity images to dually supervise the registration progress. The proposed method not only uses the intensity and segmentation information but also uses the voxelwise geometric distance information to the edges. Hence, the accurate voxelwise correspondence relationships are guaranteed both inside and outside the edges.

METHODS

The proposed dually-supervised registration method mainly includes three enhancement strategies. Firstly, we leverage the segmentation labels to construct their LSDFs to provide more geometrical information for guiding the registration process. Secondly, to calculate LSDFs, we construct an LSDF-Net, which is composed of 3D dilation layers and erosion layers. Finally, we design the dually-supervised registration network (VMLSDF ) by combining the unsupervised VoxelMorph (VM) registration network and the weakly-supervised LSDF-Net, to utilize intensity and LSDF information, respectively.

RESULTS

In this paper, experiments were then carried out on four public brain image datasets: LPBA40, HBN, OASIS1, and OASIS3. The experimental results show that the Dice similarity coefficient (DSC) and 95% Hausdorff distance (HD) of VMLSDF are higher than those of the original unsupervised VM and the dually-supervised registration network (VMseg ) using intensity images and segmentation labels. At the same time, the percentage of negative Jacobian determinant (NJD) of VMLSDF is lower than VMseg . Our code is freely available at https://github.com/1209684549/LSDF.

CONCLUSIONS

The experimental results show that LSDFs can improve the registration accuracy compared with VM and VMseg , and enhance the plausibility of the DDFs compared with VMseg .

A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial networks for medical image synthesis

Although generative adversarial networks (GANs) can produce large datasets, their limited diversity and fidelity have been recently addressed by denoising diffusion probabilistic models, which have demonstrated superiority in natural image synthesis. In this study, we introduce Medfusion, a conditional latent DDPM designed for medical image generation, and evaluate its performance against GANs, which currently represent the state-of-the-art. Medfusion was trained and compared with StyleGAN-3 using fundoscopy images from the AIROGS dataset, radiographs from the CheXpert dataset, and histopathology images from the CRCDX dataset. Based on previous studies, Progressively Growing GAN (ProGAN) and Conditional GAN (cGAN) were used as additional baselines on the CheXpert and CRCDX datasets, respectively. Medfusion exceeded GANs in terms of diversity (recall), achieving better scores of 0.40 compared to 0.19 in the AIROGS dataset, 0.41 compared to 0.02 (cGAN) and 0.24 (StyleGAN-3) in the CRMDX dataset, and 0.32 compared to 0.17 (ProGAN) and 0.08 (StyleGAN-3) in the CheXpert dataset. Furthermore, Medfusion exhibited equal or higher fidelity (precision) across all three datasets. Our study shows that Medfusion constitutes a promising alternative to GAN-based models for generating high-quality medical images, leading to improved diversity and less artifacts in the generated images.

Deep Learning for Medical Image-Based Cancer Diagnosis

(1) Background: The application of deep learning technology to realize cancer diagnosis based on medical images is one of the research hotspots in the field of artificial intelligence and computer vision. Due to the rapid development of deep learning methods, cancer diagnosis requires very high accuracy and timeliness as well as the inherent particularity and complexity of medical imaging. A comprehensive review of relevant studies is necessary to help readers better understand the current research status and ideas. (2) Methods: Five radiological images, including X-ray, ultrasound (US), computed tomography (CT), magnetic resonance imaging (MRI), positron emission computed tomography (PET), and histopathological images, are reviewed in this paper. The basic architecture of deep learning and classical pretrained models are comprehensively reviewed. In particular, advanced neural networks emerging in recent years, including transfer learning, ensemble learning (EL), graph neural network, and vision transformer (ViT), are introduced. Five overfitting prevention methods are summarized: batch normalization, dropout, weight initialization, and data augmentation. The application of deep learning technology in medical image-based cancer analysis is sorted out. (3) Results: Deep learning has achieved great success in medical image-based cancer diagnosis, showing good results in image classification, image reconstruction, image detection, image segmentation, image registration, and image synthesis. However, the lack of high-quality labeled datasets limits the role of deep learning and faces challenges in rare cancer diagnosis, multi-modal image fusion, model explainability, and generalization. (4) Conclusions: There is a need for more public standard databases for cancer. The pre-training model based on deep neural networks has the potential to be improved, and special attention should be paid to the research of multimodal data fusion and supervised paradigm. Technologies such as ViT, ensemble learning, and few-shot learning will bring surprises to cancer diagnosis based on medical images.

Improving accelerated 3D imaging in MRI-guided radiotherapy for prostate cancer using a deep learning method

PURPOSE

This study was to improve image quality for high-speed MR imaging using a deep learning method for online adaptive radiotherapy in prostate cancer. We then evaluated its benefits on image registration.

METHODS

Sixty pairs of 1.5 T MR images acquired with an MR-linac were enrolled. The data included low-speed, high-quality (LSHQ), and high-speed low-quality (HSLQ) MR images. We proposed a CycleGAN, which is based on the data augmentation technique, to learn the mapping between the HSLQ and LSHQ images and then generate synthetic LSHQ (synLSHQ) images from the HSLQ images. Five-fold cross-validation was employed to test the CycleGAN model. The normalized mean absolute error (nMAE), peak signal-to-noise ratio (PSNR), structural similarity index measurement (SSIM), and edge keeping index (EKI) were calculated to determine image quality. The Jacobian determinant value (JDV), Dice similarity coefficient (DSC), and mean distance to agreement (MDA) were used to analyze deformable registration.

RESULTS

Compared with the LSHQ, the proposed synLSHQ achieved comparable image quality and reduced imaging time by ~ 66%. Compared with the HSLQ, the synLSHQ had better image quality with improvement of 57%, 3.4%, 26.9%, and 3.6% for nMAE, SSIM, PSNR, and EKI, respectively. Furthermore, the synLSHQ enhanced registration accuracy with a superior mean JDV (6%) and preferable DSC and MDA values compared with HSLQ.

CONCLUSION

The proposed method can generate high-quality images from high-speed scanning sequences. As a result, it shows potential to shorten the scan time while ensuring the accuracy of radiotherapy.

Synthetic Attenuation Correction Maps for SPECT Imaging Using Deep Learning: A Study on Myocardial Perfusion Imaging

(1) Background: The CT-based attenuation correction of SPECT images is essential for obtaining accurate quantitative images in cardiovascular imaging. However, there are still many SPECT cameras without associated CT scanners throughout the world, especially in developing countries. Performing additional CT scans implies troublesome planning logistics and larger radiation doses for patients, making it a suboptimal solution. Deep learning (DL) offers a revolutionary way to generate complementary images for individual patients at a large scale. Hence, we aimed to generate linear attenuation coefficient maps from SPECT emission images reconstructed without attenuation correction using deep learning. (2) Methods: A total of 384 SPECT myocardial perfusion studies that used ^99mTc-sestamibi were included. A DL model based on a 2D U-Net architecture was trained using information from 312 patients. The quality of the generated synthetic attenuation correction maps (ACMs) and reconstructed emission values were evaluated using three metrics and compared to standard-of-care data using Bland-Altman plots. Finally, a quantitative evaluation of myocardial uptake was performed, followed by a semi-quantitative evaluation of myocardial perfusion. (3) Results: In a test set of 66 test patients, the ACM quality metrics were MSSIM = 0.97 ± 0.001 and NMAE = 3.08 ± 1.26 (%), and the reconstructed emission quality metrics were MSSIM = 0.99 ± 0.003 and NMAE = 0.23 ± 0.13 (%). The 95% limits of agreement (LoAs) at the voxel level for reconstructed SPECT images were: [-9.04; 9.00]%, and for the segment level, they were [-11; 10]%. The 95% LoAs for the Summed Stress Score values between the images reconstructed were [-2.8, 3.0]. When global perfusion scores were assessed, only 2 out of 66 patients showed changes in perfusion categories. (4) Conclusion: Deep learning can generate accurate attenuation correction maps from non-attenuation-corrected cardiac SPECT images. These high-quality attenuation maps are suitable for attenuation correction in myocardial perfusion SPECT imaging and could obviate the need for additional imaging in standalone SPECT scanners.