Phys Imaging Radiat Oncol. 2024 Sep 26;32:100652. doi: 10.1016/j.phro.2024.100652. eCollection 2024 Oct.

ABSTRACT

BACKGROUND AND PURPOSE: The emergence of synthetic CT (sCT) in MR-guided radiotherapy (MRgRT) represents a significant advancement, supporting MR-only workflows and online treatment adaptation. However, the lack of consensus guidelines has led to varied practices. This study reports results from a 2023 ESTRO survey aimed at defining current practices in sCT development and use.

MATERIALS AND METHODS: An survey was distributed to ESTRO members, including 98 questions across four sections on sCT algorithm generation and usage. By June 2023, 100 centers participated. The survey revealed diverse clinical experiences and roles, with primary sCT use in the pelvis (60%), brain (15%), abdomen (11%), thorax (8%), and head-and-neck (6%). sCT was mostly used for conventional fractionation treatments (68%), photon SBRT (40%), and palliative cases (28%), with limited use in proton therapy (4%).

RESULTS: Conditional GANs and GANs were the most used neural network architectures, operating mainly on 1.5 T and 3 T MRI images. Less than half used paired images for training, and only 20% performed image selection. Key MR image quality parameters included magnetic field homogeneity and spatial integrity. Half of the respondents lacked a dedicated sCT-QA program, and many did not apply sanitychecks before calculation. Selection strategies included age, weight, and metal artifacts. A strong consensus (95%) emerged for vendor neutral guidelines.

CONCLUSION: The survey highlights the need for expert-based, vendor-neutral guidelines to standardize sCT tools, metrics, and clinical protocols, ensuring effective sCT use in MR-guided radiotherapy.

PMID:39381612 | PMC:PMC11460247 | DOI:10.1016/j.phro.2024.100652

Phys Imaging Radiat Oncol. 2024 Sep 14;32:100646. doi: 10.1016/j.phro.2024.100646. eCollection 2024 Oct.

ABSTRACT

BACKGROUND AND PURPOSE: Proton Online Adaptive RadioTherapy (ProtOnART) harnesses the dosimetric advantage of protons and immediately acts upon anatomical changes. Here, we simulate the clinical application of delineation and planning within a ProtOnART-workflow for esophageal cancer. We aim to identify the most appropriate technique for autodelineation and evaluate full automation by replanning on autodelineated contours.

MATERIALS AND METHODS: We evaluated 15 patients who started treatment between 11-2022 and 01-2024, undergoing baseline and three repeat computed tomography (CT) scans in treatment position. Quantitative and qualitative evaluations compared different autodelineation methods. For Organs-at-risk (OAR) deep learning segmentation (DLS), rigid and deformable propagation from baseline to repeat CT-scans were considered. For the clinical target volume (CTV), rigid and three deformable propagation methods (default, heart as controlling structure and with focus region) were evaluated. Adaptive treatment plans with 7 mm (ATP7mm) and 3 mm (ATP3mm) setup robustness were generated using best-performing autodelineated contours. Clinical acceptance of ATPs was evaluated using goals encompassing ground-truth CTV-coverage and OAR-dose.

RESULTS: Deformation was preferred for autodelineation of heart, lungs and spinal cord. DLS was preferred for all other OARs. For CTV, deformation with focus region was the preferred method although the difference with other deformation methods was small. Nominal ATPs passed evaluation goals for 87 % of ATP7mm and 67 % of ATP3mm. This dropped to respectively 2 % and 29 % after robust evaluation. Insufficient CTV-coverage was the main reason for ATP-rejection.

CONCLUSION: Autodelineation aids a ProtOnART-workflow for esophageal cancer. Currently available tools regularly require manual annotations to generate clinically acceptable ATPs.

PMID:39381611 | PMC:PMC11460496 | DOI:10.1016/j.phro.2024.100646

Front Endocrinol (Lausanne). 2024 Sep 23;15:1386613. doi: 10.3389/fendo.2024.1386613. eCollection 2024.

ABSTRACT

INTRODUCTION: Diabetic foot ulcers (DFUs) are a severe complication among diabetic patients, often leading to amputation or even death. Early detection of infection and ischemia is essential for improving healing outcomes, but current diagnostic methods are invasive, time-consuming, and costly. There is a need for non-invasive, efficient, and affordable solutions in diabetic foot care.

METHODS: We developed DFUCare, a platform that leverages computer vision and deep learning (DL) algorithms to localize, classify, and analyze DFUs non-invasively. The platform combines CIELAB and YCbCr color space segmentation with a pre-trained YOLOv5s algorithm for wound localization. Additionally, deep-learning models were implemented to classify infection and ischemia in DFUs. The preliminary performance of the platform was tested on wound images acquired using a cell phone.

RESULTS: DFUCare achieved an F1-score of 0.80 and a mean Average Precision (mAP) of 0.861 for wound localization. For infection classification, we obtained a binary accuracy of 79.76%, while ischemic classification reached 94.81% on the validation set. The system successfully measured wound size and performed tissue color and textural analysis for a comparative assessment of macroscopic wound features. In clinical testing, DFUCare localized wounds and predicted infected and ischemic with an error rate of less than 10%, underscoring the strong performance of the platform.

DISCUSSION: DFUCare presents an innovative approach to wound care, offering a cost-effective, remote, and convenient healthcare solution. By enabling non-invasive and accurate analysis of wounds using mobile devices, this platform has the potential to revolutionize diabetic foot care and improve clinical outcomes through early detection of infection and ischemia.

PMID:39381435 | PMC:PMC11460545 | DOI:10.3389/fendo.2024.1386613

Front Nucl Med. 2023 Feb 6;2:1083245. doi: 10.3389/fnume.2022.1083245. eCollection 2022.

ABSTRACT

Prostate gland segmentation is the primary step to estimate gland volume, which aids in the prostate disease management. In this study, we present a 2D-3D convolutional neural network (CNN) ensemble that automatically segments the whole prostate gland along with the peripheral zone (PZ) (PPZ-SegNet) using a T2-weighted sequence (T2W) of Magnetic Resonance Imaging (MRI). The study used 4 different public data sets organized as Train #1 and Test #1 (independently derived from the same cohort), Test #2, Test #3 and Test #4. The prostate gland and the peripheral zone (PZ) anatomy were manually delineated with consensus read by a radiologist, except for Test #4 cohorts that had pre-marked glandular anatomy. A Bayesian hyperparameter optimization method was applied to construct the network model (PPZ-SegNet) with a training cohort (Train #1, n = 150) using a five-fold cross validation. The model evaluation was performed on an independent cohort of 283 T2W MRI prostate cases (Test #1 to #4) without any additional tuning. The data cohorts were derived from The Cancer Imaging Archives (TCIA): PROSTATEx Challenge, Prostatectomy, Repeatability studies and PROMISE12-Challenge. The segmentation performance was evaluated by computing the Dice similarity coefficient and Hausdorff distance between the estimated-deep-network identified regions and the radiologist-drawn annotations. The deep network architecture was able to segment the prostate gland anatomy with an average Dice score of 0.86 in Test #1 (n = 192), 0.79 in Test #2 (n = 26), 0.81 in Test #3 (n = 15), and 0.62 in Test #4 (n = 50). We also found the Dice coefficient improved with larger prostate volumes in 3 of the 4 test cohorts. The variation of the Dice scores from different cohorts of test images suggests the necessity of more diverse models that are inclusive of dependencies such as the gland sizes and others, which will enable us to develop a universal network for prostate and PZ segmentation. Our training and evaluation code can be accessed through the link: https://github.com/mariabaldeon/PPZ-SegNet.git.

PMID:39381408 | PMC:PMC11460296 | DOI:10.3389/fnume.2022.1083245

J Phys Chem A. 2024 Oct 8. doi: 10.1021/acs.jpca.4c04849. Online ahead of print.

ABSTRACT

Thermal resistance of energetic materials is critical due to its impact on safety and sustainability. However, developing predictive models remains challenging because of data scarcity and limited insights into quantitative structure-property relationships. In this work, a deep learning framework, named EM-thermo, was proposed to address these challenges. A data set comprising 5029 CHNO compounds, including 976 energetic compounds, was constructed to facilitate this study. EM-thermo employs molecular graphs and direct message-passing neural networks to capture structural features and predict thermal resistance. Using transfer learning, the model achieves an accuracy of approximately 97% for predicting the thermal-resistance property (decomposition temperatures above 573.15 K) in energetic compounds. The involvement of molecular descriptors improved model prediction. These findings suggest that EM-thermo is effective for correlating thermal resistance from the atom and covalent bond level, offering a promising tool for advancing molecular design and discovery in the field of energetic compounds.

PMID:39380131 | DOI:10.1021/acs.jpca.4c04849

J Orthop Surg Res. 2024 Oct 8;19(1):637. doi: 10.1186/s13018-024-05128-6.

ABSTRACT

BACKGROUND: Orthopedic surgeons use manual measurements, acetate templating, and dedicated software to determine the appropriate implant size for total knee arthroplasty (TKA). This study aimed to use deep learning (DL) to assist in deciding the femoral and tibial implant sizes without manual manipulation and to evaluate the clinical validity of the DL decision by comparing it with conventional manual procedures.

METHODS: Two types of DL were used to detect the femoral and tibial regions using the You Only Look Once algorithm model and to determine the implant size from the detected regions using convolutional neural network. An experienced surgeon predicted the implant size for 234 patient cases using manual procedures, and the DL model also predicted the implant sizes for the same cases.

RESULTS: The exact accuracies of the surgeon's template were 61.54% and 68.38% for predicting femoral and tibial implant sizes, respectively. Meanwhile, the proposed DL model reported exact accuracies of 89.32% and 90.60% for femoral and tibial implant sizes, respectively. The accuracy ± 1 levels of the surgeon and proposed DL model were 97.44% and 97.86%, respectively, for the femoral implant size and 98.72% for both the surgeon and proposed DL model for the tibial implant size.

CONCLUSION: The observed differences and higher agreement levels achieved by the proposed DL model demonstrate its potential as a valuable tool in preoperative decision-making for TKA. By providing accurate predictions of implant size, the proposed DL model has the potential to optimize implant selection, leading to improved surgical outcomes.

PMID:39380122 | DOI:10.1186/s13018-024-05128-6

Genomics Inform. 2024 Oct 8;22(1):15. doi: 10.1186/s44342-024-00019-y.

ABSTRACT

Many subtypes of SARS-CoV-2 have emerged since its early stages, with mutations showing regional and racial differences. These mutations significantly affected the infectivity and severity of the virus. This study aimed to predict the mutations that occur during the evolution of SARS-CoV-2 and identify the key characteristics for making these predictions. We collected and organized data on the lineage, date, clade, and mutations of SARS-CoV-2 from publicly available databases and processed them to predict the mutations. In addition, we utilized various artificial intelligence models to predict newly emerging mutations and created various training sets based on clade information. Using only mutation information resulted in low performance of the learning models, whereas incorporating clade differentiation resulted in high performance in machine learning models, including XGBoost (accuracy: 0.999). However, mutations fixed in the receptor-binding motif (RBM) region of Omicron resulted in decreased predictive performance. Using these models, we predicted potential mutation positions for 24C, following the recently emerged 24A and 24B clades. We identified a mutation at position Q493 in the RBM region. Our study developed effective artificial intelligence models and characteristics for predicting new mutations in continuously evolving infectious viruses.

PMID:39380083 | DOI:10.1186/s44342-024-00019-y

Genome Biol. 2024 Oct 8;25(1):258. doi: 10.1186/s13059-024-03397-2.

ABSTRACT

Precise calling of promiscuous adenosine-to-inosine RNA editing sites from transcriptomic datasets is hindered by DNA mutations and sequencing/mapping errors. Here, we present a stepwise computational framework, called DEMINING, to distinguish RNA editing and DNA mutations directly from RNA sequencing datasets, with an embedded deep learning model named DeepDDR. After transfer learning, DEMINING can also classify RNA editing sites and DNA mutations from non-primate sequencing samples. When applied in samples from acute myeloid leukemia patients, DEMINING uncovers previously underappreciated DNA mutation and RNA editing sites; some associated with the upregulated expression of host genes or the production of neoantigens.

PMID:39380061 | DOI:10.1186/s13059-024-03397-2

Radiat Oncol. 2024 Oct 8;19(1):139. doi: 10.1186/s13014-024-02524-4.

ABSTRACT

BACKGROUND: This observational study aims to establish the feasibility of using x-ray images of radio-opaque chemoembolisation deposits in patients as a method for real-time image-guided radiation therapy of hepatocellular carcinoma.

METHODS: This study will recruit 50 hepatocellular carcinoma patients who have had or will have stereotactic ablative radiation therapy and have had transarterial chemoembolisation with a radio-opaque agent. X-ray and computed tomography images of the patients will be analysed retrospectively. Additionally, a deep learning method for real-time motion tracking will be developed. We hypothesise that: (i) deep learning software can be developed that will successfully track the contrast agent mass on two thirds of cone beam computed tomography (CBCT) projection and intra-treatment images (ii), the mean and standard deviation (mm) difference in the location of the mass between ground truth and deep learning detection are ≤ 2 mm and ≤ 3 mm respectively and (iii) statistical modelling of study data will predict tracking success in 85% of trial participants.

DISCUSSION: Developing a real-time tracking method will enable increased targeting accuracy, without the need for additional invasive procedures to implant fiducial markers.

TRIAL REGISTRATION: Registered to ClinicalTrials.gov (NCT05169177) 12th October 2021.

PMID:39380004 | DOI:10.1186/s13014-024-02524-4

BMC Biol. 2024 Oct 8;22(1):225. doi: 10.1186/s12915-024-02022-9.

ABSTRACT

BACKGROUND: Homologous recombination deficiency (HRD) is recognized as a pan-cancer predictive biomarker that potentially indicates who could benefit from treatment with PARP inhibitors (PARPi). Despite its clinical significance, HRD testing is highly complex. Here, we investigated in a proof-of-concept study whether Deep Learning (DL) can predict HRD status solely based on routine hematoxylin & eosin (H&E) histology images across nine different cancer types.

METHODS: We developed a deep learning pipeline with attention-weighted multiple instance learning (attMIL) to predict HRD status from histology images. As part of our approach, we calculated a genomic scar HRD score by combining loss of heterozygosity (LOH), telomeric allelic imbalance (TAI), and large-scale state transitions (LST) from whole genome sequencing (WGS) data of n = 5209 patients across two independent cohorts. The model's effectiveness was evaluated using the area under the receiver operating characteristic curve (AUROC), focusing on its accuracy in predicting genomic HRD against a clinically recognized cutoff value.

RESULTS: Our study demonstrated the predictability of genomic HRD status in endometrial, pancreatic, and lung cancers reaching cross-validated AUROCs of 0.79, 0.58, and 0.66, respectively. These predictions generalized well to an external cohort, with AUROCs of 0.93, 0.81, and 0.73. Moreover, a breast cancer-trained image-based HRD classifier yielded an AUROC of 0.78 in the internal validation cohort and was able to predict HRD in endometrial, prostate, and pancreatic cancer with AUROCs of 0.87, 0.84, and 0.67, indicating that a shared HRD-like phenotype occurs across these tumor entities.

CONCLUSIONS: This study establishes that HRD can be directly predicted from H&E slides using attMIL, demonstrating its applicability across nine different tumor types.

PMID:39379982 | DOI:10.1186/s12915-024-02022-9

BMC Bioinformatics. 2024 Oct 8;25(1):324. doi: 10.1186/s12859-024-05915-2.

ABSTRACT

BACKGROUND: Safe drug treatment requires an understanding of the potential side effects. Identifying the frequency of drug side effects can reduce the risks associated with drug use. However, existing computational methods for predicting drug side effect frequencies heavily depend on known drug side effect frequency information. Consequently, these methods face challenges when predicting the side effect frequencies of new drugs. Although a few methods can predict the side effect frequencies of new drugs, they exhibit unreliable performance owing to the exclusion of drug-side effect relationships.

RESULTS: This study proposed CrossFeat, a model based on convolutional neural network-transformer architecture with cross-feature learning that can predict the occurrence and frequency of drug side effects for new drugs, even in the absence of information regarding drug-side effect relationships. CrossFeat facilitates the concurrent learning of drugs and side effect information within its transformer architecture. This simultaneous exchange of information enables drugs to learn about their associated side effects, while side effects concurrently acquire information about the respective drugs. Such bidirectional learning allows for the comprehensive integration of drug and side effect knowledge. Our five-fold cross-validation experiments demonstrated that CrossFeat outperforms existing studies in predicting side effect frequencies for new drugs without prior knowledge.

CONCLUSIONS: Our model offers a promising approach for predicting the drug side effect frequencies, particularly for new drugs where prior information is limited. CrossFeat's superior performance in cross-validation experiments, along with evidence from case studies and ablation experiments, highlights its effectiveness.

PMID:39379821 | DOI:10.1186/s12859-024-05915-2

Dermatol Ther (Heidelb). 2024 Oct 8. doi: 10.1007/s13555-024-01283-0. Online ahead of print.

ABSTRACT

INTRODUCTION: Developing automatic acne vulgaris grading systems based on machine learning is an expensive endeavor in terms of data acquisition. A machine learning practitioner will need to gather high-resolution pictures from a considerable number of different patients, with a well-balanced distribution between acne severity grades and potentially very tedious labeling. We developed a deep learning model to grade acne severity with respect to the Investigator's Global Assessment (IGA) scale that can be trained on low-resolution images, with pictures from a small number of different patients, a strongly imbalanced severity grade distribution and minimal labeling.

METHODS: A total of 1374 triplets of images (frontal and lateral views) from 391 different patients suffering from acne labeled with the IGA severity grade by an expert dermatologist were used to train and validate a deep learning model that predicts the IGA severity grade.

RESULTS: On the test set we obtained 66.67% accuracy with an equivalent performance for all grades despite the highly imbalanced severity grade distribution of our database. Importantly, we obtained performance on par with more tedious methods in terms of data acquisition which have the same simple labeling as ours but require either a more balanced severity grade distribution or large numbers of high-resolution images.

CONCLUSIONS: Our deep learning model demonstrated promising accuracy despite the limited data set on which it was trained, indicating its potential for further development both as an assistance tool for medical practitioners and as a way to provide patients with an immediately available and standardized acne grading tool.

TRIAL REGISTRATION: chinadrugtrials.org.cn identifier CTR20211314.

PMID:39379778 | DOI:10.1007/s13555-024-01283-0

J Cancer Res Clin Oncol. 2024 Oct 9;150(10):449. doi: 10.1007/s00432-024-05969-y.

ABSTRACT

PURPOSE: This study assesses the reliability of deep learning models based on planar whole-body bone scintigraphy for diagnosing Skull base invasion (SBI) in nasopharyngeal carcinoma (NPC) patients.

METHODS: In this multicenter study, a deep learning model was developed using data from one center with a 7:3 allocation to training and internal test sets, to diagnose SBI in patients newly diagnosed with NPC using planar whole-body bone scintigraphy. Patients were diagnosed based on a composite reference standard incorporating radiologic and follow-up data. Ten different convolutional neural network (CNN) models were applied to both whole-image and partial-image input modes to determine the optimal model for each analysis. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration, decision curve analysis (DCA), and compared with expert assessments by two nuclear medicine physicians.

RESULTS: The best-performing model using partial-body input achieved AUCs of 0.80 (95% CI: 0.73, 0.86) in the internal test set, 0.84 (95% CI: 0.77, 0.91) in the external cohort, and 0.78 (95% CI: 0.73, 0.83) in the treatment test cohort. Calibration curves and DCA confirmed the models' excellent discrimination, calibration, and potential clinical utility across internal and external datasets. The AUCs of both nuclear medicine physicians were lower than those of the best-performing deep learning model in external test set (AUC: 0.75 vs. 0.77 vs. 0.84).

CONCLUSION: Deep learning models utilizing partial-body input from planar whole-body bone scintigraphy demonstrate high discriminatory power for diagnosing SBI in NPC patients, surpassing experienced nuclear medicine physicians.

PMID:39379746 | DOI:10.1007/s00432-024-05969-y

J Cancer Res Clin Oncol. 2024 Oct 9;150(10):450. doi: 10.1007/s00432-024-05986-x.

ABSTRACT

PURPOSE: To develop and evaluate a nomogram that integrates clinical parameters with deep learning radiomics (DLR) extracted from Magnetic Resonance Imaging (MRI) data to enhance the predictive accuracy for preoperative lymph node (LN) metastasis in rectal cancer.

METHODS: A retrospective analysis was conducted on 356 patients diagnosed with rectal cancer. Of these, 286 patients were allocated to the training set, and 70 patients comprised the external validation cohort. Preprocessed T2-weighted and diffusion-weighted imaging performed preoperatively facilitated the extraction of DLR features. Five machine learning algorithms-k-nearest neighbor, light gradient boosting machine, logistic regression, random forest, and support vector machine-were utilized to develop DLR models. The most effective algorithm was identified and used to establish a clinical DLR (CDLR) nomogram specifically designed to predict LN metastasis in rectal cancer. The performance of the nomogram was evaluated using receiver operating characteristic curve analysis.

RESULTS: The logistic regression classifier demonstrated significant predictive accuracy using the DLR signature, achieving an Area Under the Curve (AUC) of 0.919 in the training cohort and 0.778 in the external validation cohort. The integrated CDLR nomogram exhibited robust predictive performance across both datasets, with AUC values of 0.921 in the training cohort and 0.818 in the external validation cohort. Notably, it outperformed both the clinical model, which had AUC values of 0.770 and 0.723 in the training and external validation cohorts, respectively, and the stand-alone DLR model.

CONCLUSION: The nomogram derived from multiparametric MRI data, referred to as the CDLR model, demonstrates strong predictive efficacy in forecasting LN metastasis in rectal cancer.

PMID:39379733 | DOI:10.1007/s00432-024-05986-x

IEEE Trans Pattern Anal Mach Intell. 2024 Oct 8;PP. doi: 10.1109/TPAMI.2024.3476075. Online ahead of print.

ABSTRACT

Achieving generalization for deep learning models has usually suffered from the bottleneck of annotated sample scarcity. As a common way of tackling this issue, few-shot learning focuses on "episodes", i.e. sampled tasks that help the model acquire generalizable knowledge onto unseen categories - better the episodes, the higher a model's generalisability. Despite extensive research, the characteristics of episodes and their potential effects are relatively less explored. A recent paper discussed that different episodes exhibit different prediction difficulties, and coined a new metric "hardness" to quantify episodes, which however is too wide-range for an arbitrary dataset and thus remains impractical for realistic applications. In this paper therefore, we for the first time conduct an algebraic analysis of the critical factors influencing episode hardness supported by experimental demonstrations, that reveal episode hardness to largely depend on classes within an episode, and importantly propose an efficient pre-sampling hardness assessment technique named Inverse-Fisher Discriminant Ratio (IFDR). This enables sampling hard episodes at the class level via class-level (cl) sampling scheme that drastically decreases quantification cost. Delving deeper, we also develop a variant called class-pair-level (cpl) sampling, which further reduces the sampling cost while guaranteeing the sampled distribution. Finally, comprehensive experiments conducted on benchmark datasets verify the efficacy of our proposed method. Codes are available at: https://github.com/PRIS-CV/class-level-sampling.

PMID:39378258 | DOI:10.1109/TPAMI.2024.3476075

IEEE Trans Cybern. 2024 Oct 8;PP. doi: 10.1109/TCYB.2024.3465213. Online ahead of print.

ABSTRACT

While highly influential in deep learning, especially in natural language processing, the Transformer model has not exhibited competitive performance in unsupervised graph representation learning (UGRL). Conventional approaches, which focus on local substructures on the graph, offer simplicity but often fall short in encapsulating comprehensive structural information of the graph. This deficiency leads to suboptimal generalization performance. To address this, we proposed the GraKerformer model, a variant of the standard Transformer architecture, to mitigate the shortfall in structural information representation and enhance the performance in UGRL. By leveraging the shortest-path graph kernel (SPGK) to weight attention scores and combining graph neural networks, the GraKerformer effectively encodes the nuanced structural information of graphs. We conducted evaluations on the benchmark datasets for graph classification to validate the superior performance of our approach.

PMID:39378254 | DOI:10.1109/TCYB.2024.3465213

IEEE Trans Image Process. 2024 Oct 8;PP. doi: 10.1109/TIP.2024.3444315. Online ahead of print.

ABSTRACT

Deep learning-based video denoising methods have achieved great performance improvements in recent years. However, the expensive computational cost arising from sophisticated network design has severely limited their applications in real-world scenarios. To address this practical weakness, we propose a multiscale spatio-temporal memory network for fast video denoising, named MSTMN, aiming at striking an improved trade-off between cost and performance. To develop an efficient and effective algorithm for video denoising, we exploit a multiscale representation based on the Gaussian-Laplacian pyramid decomposition so that the reference frame can be restored in a coarse-to-fine manner. Guided by a model-based optimization approach, we design an effective variance estimation module, an alignment error estimation module and an adaptive fusion module for each scale of the pyramid representation. For the fusion module, we employ a reconstruction recurrence strategy to incorporate local temporal information. Moreover, we propose a memory enhancement module to exploit the global spatio-temporal information. Meanwhile, the similarity computation of the spatio-temporal memory network enables the proposed network to adaptively search the valuable information at the patch level, which avoids computationally expensive motion estimation and compensation operations. Experimental results on real-world raw video datasets have demonstrated that the proposed lightweight network outperforms current state-of-the-art fast video denoising algorithms such as FastDVDnet, EMVD, and ReMoNet with fewer computational costs.

PMID:39378250 | DOI:10.1109/TIP.2024.3444315

Cardiovasc Res. 2024 Oct 1:cvae171. doi: 10.1093/cvr/cvae171. Online ahead of print.

NO ABSTRACT

PMID:39378220 | DOI:10.1093/cvr/cvae171

EJNMMI Phys. 2024 Oct 8;11(1):82. doi: 10.1186/s40658-024-00687-3.

ABSTRACT

PURPOSE: The problem of image denoising in single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) is a fundamental challenge. Although various image processing techniques have been presented, they may degrade the contrast of denoised images. The proposed idea in this study is to use a deep neural network as the denoising procedure during the iterative reconstruction process rather than the post-reconstruction phase. This method could decrease the background coefficient of variation (COV_bkg) of the final reconstructed image, which represents the amount of random noise, while improving the contrast-to-noise ratio (CNR).

METHODS: In this study, a generative adversarial network is used, where its generator is trained by a two-phase approach. In the first phase, the network is trained by a confined image region around the heart in transverse view. The second phase improves the network's generalization by tuning the network weights with the full image size as the input. The network was trained and tested by a dataset of 247 patients who underwent two immediate serially high- and low-noise SPECT-MPI.

RESULTS: Quantitative results show that compared to post-reconstruction low pass filtering and post-reconstruction deep denoising methods, our proposed method can decline the COV_bkg of the images by up to 10.28% and 12.52% and enhance the CNR by up to 54.54% and 45.82%, respectively.

CONCLUSION: The iterative deep denoising method outperforms 2D low-pass Gaussian filtering with an 8.4-mm FWHM and post-reconstruction deep denoising approaches.

PMID:39378001 | DOI:10.1186/s40658-024-00687-3

Radiology. 2024 Oct;313(1):e232749. doi: 10.1148/radiol.232749.

ABSTRACT

Background CT deep learning image reconstruction (DLIR) improves image quality by reducing noise compared with adaptive statistical iterative reconstruction-V (ASIR-V). However, objective assessment of low-contrast lesion detectability is lacking. Purpose To investigate low-contrast detectability of hypoattenuating liver lesions on CT scans reconstructed with DLIR compared with CT scans reconstructed with ASIR-V in a patient and a phantom study. Materials and Methods This single-center retrospective study included patients undergoing portal venous phase abdominal CT between February and May 2021 and a low-contrast-resolution phantom scanned with the same protocol. Four reconstructions (ASIR-V at 40% strength [ASIR-V 40] and DLIR at three strengths) were generated. Five radiologists qualitatively assessed the images using the five-point Likert scale for image quality, lesion diagnostic confidence, conspicuity, and small lesion (≤1 cm) visibility. Up to two key lesions per patient, confirmed at histopathologic testing or at prior or follow-up imaging studies, were included. Lesion-to-background contrast-to-noise ratio was calculated. Interreader variability was analyzed. Intergroup qualitative and quantitative metrics were compared between DLIR and ASIR-V 40 using proportional odds logistic regression models. Results Eighty-six liver lesions (mean size, 15 mm ± 9.5 [SD]) in 50 patients (median age, 62 years [IQR, 57-73 years]; 27 [54%] female patients) were included. Differences were not detected for various qualitative low-contrast detectability metrics between ASIR-V 40 and DLIR (P > .05). Quantitatively, medium-strength DLIR and high-strength DLIR yielded higher lesion-to-background contrast-to-noise ratios than ASIR-V 40 (medium-strength DLIR vs ASIR-V 40: odds ratio [OR], 1.96 [95% CI: 1.65, 2.33]; high-strength DLIR vs ASIR-V 40: OR, 5.36 [95% CI: 3.68, 7.82]; P < .001). Low-contrast lesion attenuation was reduced by 2.8-3.6 HU with DLIR. Interreader agreement was moderate to very good for the qualitative metrics. Subgroup analysis based on lesion size of larger than 1 cm and 1 cm or smaller yielded similar results (P > .05). Qualitatively, phantom study results were similar to those in patients (P > .05). Conclusion The detectability of low-contrast liver lesions was similar on CT scans reconstructed with low-, medium-, and high-strength DLIR and ASIR-V 40 in both patient and phantom studies. Lesion-to-background contrast-to-noise ratios were higher for DLIR medium- and high-strength reconstructions compared with ASIR-V 40. © RSNA, 2024 Supplemental material is available for this article.

PMID:39377679 | DOI:10.1148/radiol.232749