Journal of Digital Imaging最新文献

To explore the value of CT-based radiomics model in the differential diagnosis of benign ovarian tumors (BeOTs), borderline ovarian tumors (BOTs), and early malignant ovarian tumors (eMOTs). The retrospective research was conducted with pathologically confirmed 258 ovarian tumor patients from January 2014 to February 2021. The patients were randomly allocated to a training cohort (n = 198) and a test cohort (n = 60). By providing a three-dimensional (3D) characterization of the volume of interest (VOI) at the maximum level of images, 4238 radiomic features were extracted from the VOI per patient. The Wilcoxon–Mann–Whitney (WMW) test, least absolute shrinkage and selection operator (LASSO), and support vector machine (SVM) were employed to select the radiomic features. Five machine learning (ML) algorithms were applied to construct three-class diagnostic models. Leave-one-out cross-validation (LOOCV) was implemented to evaluate the performance of the radiomics models. The test cohort was used to verify the generalization ability of the radiomics models. The receiver-operating characteristic (ROC) was used to evaluate diagnostic performance of radiomics model. Global and discrimination performance of five models was evaluated by average area under the ROC curve (AUC). The average ROC indicated that random forest (RF) diagnostic model in training cohort demonstrated the best diagnostic performance (micro/macro average AUC, 0.98/0.99), which was then confirmed with by LOOCV (micro/macro average AUC, 0.89/0.88) and external validation (test cohort) (micro/macro average AUC, 0.81/0.79). Our proposed CT-based radiomics diagnostic models may effectively assist in preoperatively differentiating BeOTs, BOTs, and eMOTs.

Pretreatment patient-specific quality assurance (prePSQA) is conducted to confirm the accuracy of the radiotherapy dose delivered. However, the process of prePSQA measurement is time consuming and exacerbates the workload for medical physicists. The purpose of this work is to propose a novel deep learning (DL) network to improve the accuracy and efficiency of prePSQA. A modified invertible and variable augmented network was developed to predict the three-dimensional (3D) measurement-guided dose (MDose) distribution of 300 cancer patients who underwent volumetric modulated arc therapy (VMAT) between 2018 and 2021, in which 240 cases were randomly selected for training, and 60 for testing. For simplicity, the present approach was termed as “IVPSQA.” The input data include CT images, radiotherapy dose exported from the treatment planning system, and MDose distribution extracted from the verification system. Adam algorithm was used for first-order gradient-based optimization of stochastic objective functions. The IVPSQA model obtained high-quality 3D prePSQA dose distribution maps in head and neck, chest, and abdomen cases, and outperformed the existing U-Net-based prediction approaches in terms of dose difference maps and horizontal profiles comparison. Moreover, quantitative evaluation metrics including SSIM, MSE, and MAE demonstrated that the proposed approach achieved a good agreement with ground truth and yield promising gains over other advanced methods. This study presented the first work on predicting 3D prePSQA dose distribution by using the IVPSQA model. The proposed method could be taken as a clinical guidance tool and help medical physicists to reduce the measurement work of prePSQA.

Deep learning (DL) has recently attracted attention for data processing in positron emission tomography (PET). Attenuation correction (AC) without computed tomography (CT) data is one of the interests. Here, we present, to our knowledge, the first attempt to generate an attenuation map of the human head via Sim2Real DL-based tissue composition estimation from model training using only the simulated PET dataset. The DL model accepts a two-dimensional non-attenuation-corrected PET image as input and outputs a four-channel tissue-composition map of soft tissue, bone, cavity, and background. Then, an attenuation map is generated by a linear combination of the tissue composition maps and, finally, used as input for scatter+random estimation and as an initial estimate for attenuation map reconstruction by the maximum likelihood attenuation correction factor (MLACF), i.e., the DL estimate is refined by the MLACF. Preliminary results using clinical brain PET data showed that the proposed DL model tended to estimate anatomical details inaccurately, especially in the neck-side slices. However, it succeeded in estimating overall anatomical structures, and the PET quantitative accuracy with DL-based AC was comparable to that with CT-based AC. Thus, the proposed DL-based approach combined with the MLACF is also a promising CT-less AC approach.

Deep stromal invasion is an important pathological factor associated with the treatments and prognosis of cervical cancer patients. Accurate determination of deep stromal invasion before radical hysterectomy (RH) is of great value for early clinical treatment decision-making and improving the prognosis of these patients. Machine learning is gradually applied in the construction of clinical models to improve the accuracy of clinical diagnosis or prediction, but whether machine learning can improve the preoperative diagnosis accuracy of deep stromal invasion in patients with cervical cancer was still unclear. This cross-sectional study was to construct three preoperative diagnostic models for deep stromal invasion in patients with early cervical cancer based on clinical, radiomics, and clinical combined radiomics data using the machine learning method. We enrolled 229 patients with early cervical cancer receiving RH combined with pelvic lymph node dissection (PLND). The least absolute shrinkage and selection operator (LASSO) and the fivefold cross-validation were applied to screen out radiomics features. Univariate and multivariate logistic regression analyses were applied to identify clinical predictors. All subjects were divided into the training set (n = 160) and testing set (n = 69) at a ratio of 7:3. Three light gradient boosting machine (LightGBM) models were constructed in the training set and verified in the testing set. The radiomics features were statistically different between deep stromal invasion < 1/3 group and deep stromal invasion ≥ 1/3 group. In the training set, the area under the curve (AUC) of the prediction model based on radiomics features was 0.951 (95% confidence interval (CI) 0.922–0.980), the AUC of the prediction model based on clinical predictors was 0.769 (95% CI 0.703–0.835), and the AUC of the prediction model based on radiomics features and clinical predictors was 0.969 (95% CI 0.947–0.990). The AUC of the prediction model based on radiomics features and clinical predictors was 0.914 (95% CI 0.848–0.980) in the testing set. The prediction model for deep stromal invasion in patients with early cervical cancer based on clinical and radiomics data exhibited good predictive performance with an AUC of 0.969, which might help the clinicians early identify patients with high risk of deep stromal invasion and provide timely interventions.

Pharmacokinetic (PK) parameters, revealing changes in the tumor microenvironment, are related to the pathological information of breast cancer. Tracer kinetic models (e.g., Tofts-Kety model) with a nonlinear least square solver are commonly used to estimate PK parameters. However, the method is sensitive to noise in images. To relieve the effects of noise, a deconvolution (DEC) method, which was validated on synthetic concentration–time series, was proposed to accurately calculate PK parameters from breast dynamic contrast-enhanced magnetic resonance imaging. A time-to-peak-based tumor partitioning method was used to divide the whole tumor into three tumor subregions with different kinetic patterns. Radiomic features were calculated from the tumor subregion and whole tumor-based PK parameter maps. The optimal features determined by the fivefold cross-validation method were used to build random forest classifiers to predict molecular subtypes, Ki-67, and tumor grade. The diagnostic performance evaluated by the area under the receiver operating characteristic curve (AUC) was compared between the subregion and whole tumor-based PK parameters. The results showed that the DEC method obtained more accurate PK parameters than the Tofts method. Moreover, the results showed that the subregion-based K^trans (best AUCs = 0.8319, 0.7032, 0.7132, 0.7490, 0.8074, and 0.6950) achieved a better diagnostic performance than the whole tumor-based K^trans (AUCs = 0.8222, 0.6970, 0.6511, 0.7109, 0.7620, and 0.5894) for molecular subtypes, Ki-67, and tumor grade. These findings indicate that DEC-based K^trans in the subregion has the potential to accurately predict molecular subtypes, Ki-67, and tumor grade.

Knowledge of input blood to the brain, which is represented as total cerebral blood flow (tCBF), is important in evaluating brain health. Phase-contrast (PC) magnetic resonance imaging (MRI) enables blood velocity mapping, allowing for noninvasive measurements of tCBF. In the procedure, manual selection of brain-feeding arteries is an essential step, but is time-consuming and often subjective. Thus, the purpose of this work was to develop and validate a deep learning (DL)-based technique for automated tCBF quantifications. To enhance the DL segmentation performance on arterial blood vessels, in the preprocessing step magnitude and phase images of PC MRI were multiplied several times. Thereafter, a U-Net was trained on 218 images for three-class segmentation. Network performance was evaluated in terms of the Dice coefficient and the intersection-over-union (IoU) on 40 test images, and additionally, on externally acquired 20 datasets. Finally, tCBF was calculated from the DL-predicted vessel segmentation maps, and its accuracy was statistically assessed with the correlation of determination (R²), the intraclass correlation coefficient (ICC), paired t-tests, and Bland-Altman analysis, in comparison to manually derived values. Overall, the DL segmentation network provided accurate labeling of arterial blood vessels for both internal (Dice=0.92, IoU=0.86) and external (Dice=0.90, IoU=0.82) tests. Furthermore, statistical analyses for tCBF estimates revealed good agreement between automated versus manual quantifications in both internal (R²=0.85, ICC=0.91, p=0.52) and external (R²=0.88, ICC=0.93, p=0.88) test groups. The results suggest feasibility of a simple and automated protocol for quantifying tCBF from neck PC MRI and deep learning.

Abstract

In the field of medicine, rapidly and accurately segmenting organs in medical images is a crucial application of computer technology. This paper introduces a feature map module, Strength Attention Area Signed Distance Map (SAA-SDM), based on the principal component analysis (PCA) principle. The module is designed to accelerate neural networks’ convergence speed in rapidly achieving high precision. SAA-SDM provides the neural network with confidence information regarding the target and background, similar to the signed distance map (SDM), thereby enhancing the network’s understanding of semantic information related to the target. Furthermore, this paper presents a training scheme tailored for the module, aiming to achieve finer segmentation and improved generalization performance. Validation of our approach is carried out using TRUS and chest X-ray datasets. Experimental results demonstrate that our method significantly enhances neural networks’ convergence speed and precision. For instance, the convergence speed of UNet and UNET + + is improved by more than 30%. Moreover, Segformer achieves an increase of over 6% and 3% in mIoU (mean Intersection over Union) on two test datasets without requiring pre-trained parameters. Our approach reduces the time and resource costs associated with training neural networks for organ segmentation tasks while effectively guiding the network to achieve meaningful learning even without pre-trained parameters. 

We aimed to develop machine learning (ML)-based algorithms to assist physicians in ultrasound-guided localization of cricoid cartilage (CC) and thyroid cartilage (TC) in cricothyroidotomy. Adult female volunteers were prospectively recruited from two hospitals between September and December, 2020. Ultrasonographic images were collected via a modified longitudinal technique. You Only Look Once (YOLOv5s), Faster Regions with Convolutional Neural Network features (Faster R-CNN), and Single Shot Detector (SSD) were selected as the model architectures. A total of 488 women (mean age: 36.0 years) participated in the study, contributing to a total of 292,053 frames of ultrasonographic images. The derived ML-based algorithms demonstrated excellent discriminative performance for the presence of CC (area under the receiver operating characteristic curve [AUC]: YOLOv5s, 0.989, 95% confidence interval [CI]: 0.982–0.994; Faster R-CNN, 0.986, 95% CI: 0.980–0.991; SSD, 0.968, 95% CI: 0.956–0.977) and TC (AUC: YOLOv5s, 0.989, 95% CI: 0.977–0.997; Faster R-CNN, 0.981, 95% CI: 0.965–0.991; SSD, 0.982, 95% CI: 0.973–0.990). Furthermore, in the frames where the model could correctly indicate the presence of CC or TC, it also accurately localized CC (intersection-over-union: YOLOv5s, 0.753, 95% CI: 0.739–0.765; Faster R-CNN, 0.720, 95% CI: 0.709–0.732; SSD, 0.739, 95% CI: 0.726–0.751) or TC (intersection-over-union: YOLOv5s, 0.739, 95% CI: 0.722–0.755; Faster R-CNN, 0.709, 95% CI: 0.687–0.730; SSD, 0.713, 95% CI: 0.695–0.730). The ML-based algorithms could identify anatomical landmarks for cricothyroidotomy in adult females with favorable discriminative and localization performance. Further studies are warranted to transfer this algorithm to hand-held portable ultrasound devices for clinical use.

This paper presents a fully automated pipeline using a sparse convolutional autoencoder for quality control (QC) of affine registrations in large-scale T1-weighted (T1w) and T2-weighted (T2w) magnetic resonance imaging (MRI) studies. Here, a customized 3D convolutional encoder-decoder (autoencoder) framework is proposed and the network is trained in a fully unsupervised manner. For cross-validating the proposed model, we used 1000 correctly aligned MRI images of the human connectome project young adult (HCP-YA) dataset. We proposed that the quality of the registration is proportional to the reconstruction error of the autoencoder. Further, to make this method applicable to unseen datasets, we have proposed dataset-specific optimal threshold calculation (using the reconstruction error) from ROC analysis that requires a subset of the correctly aligned and artificially generated misalignments specific to that dataset. The calculated optimum threshold is used for testing the quality of remaining affine registrations from the corresponding datasets. The proposed framework was tested on four unseen datasets from autism brain imaging data exchange (ABIDE I, 215 subjects), information eXtraction from images (IXI, 577 subjects), Open Access Series of Imaging Studies (OASIS4, 646 subjects), and “Food and Brain” study (77 subjects). The framework has achieved excellent performance for T1w and T2w affine registrations with an accuracy of 100% for HCP-YA. Further, we evaluated the generality of the model on four unseen datasets and obtained accuracies of 81.81% for ABIDE I (only T1w), 93.45% (T1w) and 81.75% (T2w) for OASIS4, and 92.59% for “Food and Brain” study (only T1w) and in the range 88–97% for IXI (for both T1w and T2w and stratified concerning scanner vendor and magnetic field strengths). Moreover, the real failures from “Food and Brain” and OASIS4 datasets were detected with sensitivities of 100% and 80% for T1w and T2w, respectively. In addition, AUCs of > 0.88 in all scenarios were obtained during threshold calculation on the four test sets.

Cervical cancer is a significant health problem worldwide, and early detection and treatment are critical to improving patient outcomes. To address this challenge, a deep learning (DL)-based cervical classification system is proposed using 3D convolutional neural network and Vision Transformer (ViT) module. The proposed model leverages the capability of 3D CNN to extract spatiotemporal features from cervical images and employs the ViT model to capture and learn complex feature representations. The model consists of an input layer that receives cervical images, followed by a 3D convolution block, which extracts features from the images. The feature maps generated are down-sampled using max-pooling block to eliminate redundant information and preserve important features. Four Vision Transformer models are employed to extract efficient feature maps of different levels of abstraction. The output of each Vision Transformer model is an efficient set of feature maps that captures spatiotemporal information at a specific level of abstraction. The feature maps generated by the Vision Transformer models are then supplied into the 3D feature pyramid network (FPN) module for feature concatenation. The 3D squeeze-and-excitation (SE) block is employed to obtain efficient feature maps that recalibrate the feature responses of the network based on the interdependencies between different feature maps, thereby improving the discriminative power of the model. At last, dimension minimization of feature maps is executed using 3D average pooling layer. Its output is then fed into a kernel extreme learning machine (KELM) for classification into one of the five classes. The KELM uses radial basis kernel function (RBF) for mapping features in high-dimensional feature space and classifying the input samples. The superiority of the proposed model is known using simulation results, achieving an accuracy of 98.6%, demonstrating its potential as an effective tool for cervical cancer classification. Also, it can be used as a diagnostic supportive tool to assist medical experts in accurately identifying cervical cancer in patients.