External Test Set Research Articles

Development and validation of a multimodal model in predicting severe acute pancreatitis based on radiomics and deep learning

ObjectiveAim to establish a multimodal model for predicting severe acute pancreatitis (SAP) using machine learning (ML) and deep learning (DL). MethodsIn this multicentre retrospective study, patients diagnosed with acute pancreatitis at admission were enrolled from January 2017 to December 2021. Clinical information within 24 h and CT scans within 72 h of admission were collected. First, we trained Model α based on clinical features selected by least absolute shrinkage and selection operator analysis. Second, radiomics features were extracted from 3D-CT scans and Model β was developed on the features after dimensionality reduction using principal component analysis. Third, Model γ was trained on 2D-CT images. Lastly, a multimodal model, namely PrismSAP, was constructed based on aforementioned features in the training set. The predictive accuracy of PrismSAP was verified in the validation and internal test sets and further validated in the external test set. Model performance was evaluated using area under the curve (AUC), accuracy, sensitivity, specificity, recall, precision and F1-score. ResultsA total of 1,221 eligible patients were randomly split into a training set (n = 864), a validation set (n = 209) and an internal test set (n = 148). Data of 266 patients were for external testing. In the external test set, PrismSAP performed best with the highest AUC of 0.916 (0.873–0.960) among all models [Model α: 0.709 (0.618–0.800); Model β: 0.749 (0.675–0.824); Model γ: 0.687 (0.592–0.782); MCTSI: 0.778 (0.698–0.857); RANSON: 0.642 (0.559–0.725); BISAP: 0.751 (0.668–0.833); SABP: 0.710 (0.621–0.798)]. ConclusionThe proposed multimodal model outperformed any single-modality models and traditional scoring systems.

Deep learning for automatic volumetric segmentation of left ventricular myocardium and ischaemic scar from multi-slice late gadolinium enhancement cardiovascular magnetic resonance.

This study details application of deep learning for automatic volumetric segmentation of left ventricular (LV) myocardium and scar and automated quantification of myocardial ischaemic scar burden from late gadolinium enhancement cardiovascular magnetic resonance (LGE-CMR). We included 501 images and manual segmentations of short-axis LGE-CMR from over 20 multinational sites, from which 377 studies were used for training and 124 studies from unique participants for internal validation. A third test set of 52 images was used for external evaluation. Three models, U-Net, Cascaded U-Net, and U-Net++, were trained with a novel adaptive weighted categorical cross-entropy loss function. Model performance was evaluated using concordance correlation coefficients (CCCs) for LV mass and per cent myocardial scar burden. Cascaded U-Net was found to be the best model for the quantification of LV mass and scar percentage. The model exhibited a mean difference of -5 ± 23 g for LV mass, -0.4 ± 11.2 g for scar mass, and -0.8 ± 7% for per cent scar. CCC were 0.87, 0.77, and 0.78 for LV mass, scar mass, and per cent scar burden, respectively, in the internal validation set and 0.75, 0.71, and 0.69, respectively, in the external test set. For segmental scar mass, CCC was 0.74 for apical scar, 0.91 for mid-ventricular scar, and 0.73 for basal scar, demonstrating moderate to strong agreement. We successfully trained a convolutional neural network for volumetric segmentation and analysis of LV scar burden from LGE-CMR images in a large, multinational cohort of participants with ischaemic scar.

Multi-center application of a convolutional neural network for preoperative detection of cavernous sinus invasion in pituitary adenomas.

Cavernous sinus invasion (CSI) plays a pivotal role in determining management in pituitary adenomas. The study aimed to develop a Convolutional Neural Network (CNN) model to diagnose CSI in multiple centers. A total of 729 cases were retrospectively obtained in five medical centers with (n = 543) or without CSI (n = 186) from January 2011 to December 2021. The CNN model was trained using T1-enhanced MRI from two pituitary centers of excellence (n = 647). The other three municipal centers (n = 82) as the external testing set were imported to evaluate the model performance. The area-under-the-receiver-operating-characteristic-curve values (AUC-ROC) analyses were employed to evaluate predicted performance. Gradient-weighted class activation mapping (Grad-CAM) was used to determine models' regions of interest. The CNN model achieved high diagnostic accuracy (0.89) in identifying CSI in the external testing set, with an AUC-ROC value of 0.92 (95% CI, 0.88-0.97), better than CSI clinical predictor of diameter (AUC-ROC: 0.75), length (AUC-ROC: 0.80), and the three kinds of dichotomizations of the Knosp grading system (AUC-ROC: 0.70-0.82). In cases with Knosp grade 3A (n = 24, CSI rate, 0.35), the accuracy the model accounted for 0.78, with sensitivity and specificity values of 0.72 and 0.78, respectively. According to the Grad-CAM results, the views of the model were confirmed around the sellar region with CSI. The deep learning model is capable of accurately identifying CSI and satisfactorily able to localize CSI in multicenters.

Nondestructive discrimination of Potentilla anserina L. from different production areas based on near-infrared spectroscopy.

The traditional Chinese medicine (TCM) Potentilla anserina L. can use both as food and medicine. At present, the market mainly depends on experience to identify the species and determine the production areas of P. anserina. To ensure the quality of P. anserina, it is essential to improve the level of quality control. We aimed to establish a rapid and nondestructive discrimination model to identify P. anserina from different production areas by near-infrared spectroscopy. The spectra of complete P. anserina medicinal materials and their powder of the same variety from four production areas were collected, and principal component analysis discriminant analysis and partial least squares discriminant analysis (PLS-DA) were conducted based on different pretreatment methods and band selection methods. Then, the spectra of complete medicinal materials were converted into the spectra of medicinal powder for nondestructive identification. The correct recognition rate (CRR) of the PLS-DA discriminant model was the best after spectral preprocessing using autoscaling and competitive adaptive reweighted sampling for band selection. The CRRs of the calibration set and validation set were 100%, the CRRs of the external test set were 95%, 90%, 82%, and 88%, respectively, and the CRRs of the transfer external test set were 84%, 80%, 82%, and 86%, respectively. We realized the nondestructive and effective identification of P. anserina from different origins and laid a foundation for the industrialization and upgrading of TCM.

Automated MRI liver segmentation for anatomical segmentation, liver volumetry, and the extraction of radiomics.

To develop and evaluate a deep convolutional neural network (DCNN) for automated liver segmentation, volumetry, and radiomic feature extraction on contrast-enhanced portal venous phase magnetic resonance imaging (MRI). This retrospective study included hepatocellular carcinoma patients from an institutional database with portal venous MRI. After manual segmentation, the data was randomly split into independent training, validation, and internal testing sets. From a collaborating institution, de-identified scans were used for external testing. The public LiverHccSeg dataset was used for further external validation. A 3D DCNN was trained to automatically segment the liver. Segmentation accuracy was quantified by theDice similarity coefficient (DSC) with respect to manual segmentation. A Mann-Whitney U test was used to compare the internal and external test sets. Agreement of volumetry and radiomic features was assessed using the intraclass correlation coefficient (ICC). In total, 470 patients met the inclusion criteria (63.9±8.2 years; 376 males) and 20 patients were used for external validation (41±12 years; 13 males). DSC segmentation accuracy of the DCNN was similarly high between the internal (0.97±0.01) and external (0.96±0.03) test sets (p=0.28) and demonstrated robust segmentation performance on public testing (0.93±0.03). Agreement of liver volumetry was satisfactory in the internal (ICC, 0.99), external (ICC, 0.97), and public (ICC, 0.85) test sets. Radiomic features demonstrated excellent agreement in the internal (mean ICC, 0.98±0.04), external (mean ICC, 0.94±0.10), and public (mean ICC, 0.91±0.09) datasets. Automated liver segmentation yields robust and generalizable segmentation performance on MRI data and can be used for volumetry and radiomic feature extraction. Liver volumetry, anatomic localization, and extraction of quantitative imaging biomarkers require accurate segmentation, but manual segmentation is time-consuming. A deep convolutional neural network demonstrates fast and accurate segmentation performance on T1-weighted portal venous MRI. • This deep convolutional neural network yields robust and generalizable liver segmentation performance on internal, external, and public testing data. • Automated liver volumetry demonstrated excellent agreement with manual volumetry. • Automated liver segmentations can be used for robust and reproducible radiomic feature extraction.

Rapid determination of alprazolam in gin tonic cocktails based on the coupling of IR spectroscopy and chemometrics: A feasibility study

“Drug-facilitated sexual assault" (DFSA) is a sexual assault perpetrated against a person rendered unconscious by a substance that changes her/his physical and/or mental condition, such as ethanol or drugs. Several active pharmaceutical ingredients, whether used alone or with alcoholic beverages, can produce anterograde amnesia and loss of inhibition. The most common pharmaceuticals found in DFSAs are GHB (-hydroxybutyric acid), benzodiazepines (Valium, Xanax, or Roipnol), antidepressants (Venlafaxine), muscle relaxants (cyclobenzaprine), antihistamines, sleeping pills (diphenhydramines), hallucinogens, and opioids. Biological samples are typically examined in cases of suspected DFSA; however, occasionally, samples are sent to labs a long period after being collected, jeopardizing the accuracy of the analysis. As a result, in recent years, the focus has shifted to directly detecting the presence of drugs in alcoholic beverages. In light of this, the purpose of the current study is to build a FT-IR-based approach for the determination of alprazolam in a common long drink (gin and tonic). To achieve this goal, pure (Class Pure) and spiked gin tonics (Class Spiked) were analyzed by Fourier-transform infrared spectroscopy (FT-IR). Afterward, two classifiers were used: Sequential preprocessing through ORThogonalization Linear Discriminant Analysis (SPORT-LDA) and Soft Independent Modeling of Class Analogies (SIMCA). Both approaches provided good results: SPORT-LDA achieved a 95% and a 98% accuracy rate (on the external test set of samples) for spiked and pure cocktails, respectively. This corresponds to the misclassification of 5 spiked and 1 pure drinks. The SIMCA model of class pure achieved 98.2% and 91.7% of specificity and sensitivity, respectively, coinciding with 55 pure samples (over 60) correctly accepted and 2 (over 110) erroneously rejected by the model. In conclusion, the SIMCA model of class pure seems preferable, because it minimizes the type II error. Eventually, the study was circumscribed to the spiked cocktails and a novel SPORT model was used to quantify alprazolam in spiked cocktails. This provided noteworthy results, in fact, it led to a Root Mean Square Error in Prediction (RMSEP) of 0.95, and a R2pred of 0.98.

A Transvaginal Ultrasound-Based Deep Learning Model for the Noninvasive Diagnosis of Myometrial Invasion in Patients with Endometrial Cancer: Comparison with Radiologists

ObjectiveThis study aimed to determine the feasibility of using the deep learning (DL) method to determine the degree (whether MI > 50%) of myometrial invasion (MI) in patients with endometrial cancer (EC) based on ultrasound (US) images. MethodsFrom September 2017 to April 2023, 1289 US images of 604 patients with EC who underwent surgical resection at Center 1, Center 2 or Center 3 were obtained and divided into a training set and an internal validation set. Ninety-five patients from Center 4 and Center 5 were randomly selected as the external testing set according to the same criteria as those for the primary cohort. This study evaluated three DL models trained on the training set and tested them on the validation and testing sets. The models’ performance was analyzed based on accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (AUC), and the performance of the models was subsequently compared with that of 15 radiologists. ResultsIn the final clinical diagnosis of MI in patients with EC, EfficientNet-B6 showed the best performance in the testing set in terms of area under the curve (AUC) [0.814, 95% CI (0.746-0.882]; accuracy [0.802, 95% CI (0.733-0.855]; sensitivity [0.623]; specificity [0.879]; positive likelihood ratio [PLR] [6.750]; and negative likelihood ratio [NLR] [0.389]. The diagnostic efficacy of EfficientNet-B6 was significantly better than that of the 15 radiologists, with an average diagnostic accuracy of 0.681, average AUC of 0.678, AUC of the best performance of 0.739, accuracy of 0.716, sensitivity of 0.806, specificity 0.672, PLR2.457, and NLR 0.289. ConclusionBased on the preoperative US images of patients with EC, the DL model can accurately determine the degree of endometrial MI; the performance of this model is significantly better than that of radiologists, and it can effectively assist in clinical treatment decisions.

Quantitative Predictive Studies of Multiple Biological Activities of TRPV1 Modulators.

TRPV1 channel agonists and antagonists, which have powerful analgesic effects without the addictive qualities associated with traditional analgesics, have become a focus area for the development of novel analgesics. In this study, quantitative structure-activity relationship (QSAR) models for three bioactive endpoints (Ki, IC50, and EC50) were successfully constructed using four machine learning algorithms: SVM, Bagging, GBDT, and XGBoost. These models were based on 2922 TRPV1 modulators and incorporated four types of molecular descriptors: Daylight, E-state, ECFP4, and MACCS. After the rigorous five-fold cross-validation and external test set validation, the optimal models for the three endpoints were obtained. For the Ki endpoint, the Bagging-ECFP4 model had a Q2 value of 0.778 and an R2 value of 0.780. For the IC50 endpoint, the XGBoost-ECFP4 model had a Q2 value of 0.806 and an R2 value of 0.784. For the EC50 endpoint, the SVM-Daylight model had a Q2 value of 0.784 and an R2 value of 0.809. These results demonstrate that the constructed models exhibit good predictive performance. In addition, based on the model feature importance analysis, the influence between substructure and biological activity was also explored, which can provide important theoretical guidance for the efficient virtual screening and structural optimization of novel TRPV1 analgesics. And subsequent studies on novel TRPV1 modulators will be based on the feature substructures of the three endpoints.

Generative models for color normalization in digital pathology and dermatology: Advancing the learning paradigm

Color medical images introduce an additional confounding factor compared to conventional grayscale medical images: color variability. This variability can lead to inconsistent evaluation by clinicians and the misinterpretation or suboptimal learning process of automatic quantitative algorithms. To mitigate the potential negative consequences of color variability, several color normalization strategies have been developed, proving to be effective in standardizing image appearance. In this paper, we present a novel paradigm for color normalization using generative adversarial networks (GANs). Our method focuses on standardizing images in the field of digital pathology (stain normalization) and dermatology (color constancy), where high color variability is consistently observed. Specifically, we formulate the color normalization task as an image-to-image translation problem, ensuring a pixel-to-pixel correspondence between the original and normalized images. Our approach outperforms existing state-of-the-art methods in both the digital pathology and dermatology fields. Extensive validation using public datasets demonstrate the effectiveness of our color normalization results on entirely external test sets. Our framework exhibits strong generalization capability on unseen data, making it suitable for inclusion in the pipeline of automatic quantitative algorithms to reduce color variability and improve segmentation and/or classification performance. Lastly, we provide the source code of our models to encourage open science.

Feasibility of a deep learning artificial intelligence model for the diagnosis of pediatric ileocolic intussusception with grayscale ultrasonography.

This study explored the feasibility of utilizing a deep learning artificial intelligence (AI) model to detect ileocolic intussusception on grayscale ultrasound images. This retrospective observational study incorporated ultrasound images of children who underwent emergency ultrasonography for suspected ileocolic intussusception. After excluding video clips, Doppler images, and annotated images, 40,765 images from two tertiary hospitals were included (positive-to-negative ratio: hospital A, 2,775:35,373; hospital B, 140:2,477). Images from hospital A were split into a training set, a tuning set, and an internal test set (ITS) at a ratio of 7:1.5:1.5. Images from hospital B comprised an external test set (ETS). For each image indicating intussusception, two radiologists provided a bounding box as the ground-truth label. If intussusception was suspected in the input image, the model generated a bounding box with a confidence score (0-1) at the estimated lesion location. Average precision (AP) was used to evaluate overall model performance. The performance of practical thresholds for the modelgenerated confidence score, as determined from the ITS, was verified using the ETS. The AP values for the ITS and ETS were 0.952 and 0.936, respectively. Two confidence thresholds, CTopt and CTprecision, were set at 0.557 and 0.790, respectively. For the ETS, the perimage precision and recall were 95.7% and 80.0% with CTopt, and 98.4% and 44.3% with CTprecision. For per-patient diagnosis, the sensitivity and specificity were 100.0% and 97.1% with CTopt, and 100.0% and 99.0% with CTprecision. The average number of false positives per patient was 0.04 with CTopt and 0.01 for CTprecision. The feasibility of using an AI model to diagnose ileocolic intussusception on ultrasonography was demonstrated. However, further study involving bias-free data is warranted for robust clinical validation.

Deep learning-assisted diagnosis of benign and malignant parotid gland tumors based on automatic segmentation of ultrasound images: a multicenter retrospective study.

To construct deep learning-assisted diagnosis models based on automatic segmentation of ultrasound images to facilitate radiologists in differentiating benign and malignant parotid tumors. A total of 582 patients histopathologically diagnosed with PGTs were retrospectively recruited from 4 centers, and their data were collected for analysis. The radiomics features of six deep learning models (ResNet18, Inception_v3 etc) were analyzed based on the ultrasound images that were obtained under the best automatic segmentation model (Deeplabv3, UNet++, and UNet). The performance of three physicians was compared when the optimal model was used and not. The Net Reclassification Index (NRI) and Integrated Discrimination Improvement (IDI) were utilized to evaluate the clinical benefit of the optimal model. The Deeplabv3 model performed optimally in terms of automatic segmentation. The ResNet18 deep learning model had the best prediction performance, with an area under the receiver-operating characteristic curve of 0.808 (0.694-0.923), 0.809 (0.712-0.906), and 0.812 (0.680-0.944) in the internal test set and external test sets 1 and 2, respectively. Meanwhile, the optimal model-assisted clinical and overall benefits were markedly enhanced for two out of three radiologists (in internal validation set, NRI: 0.259 and 0.213 [p = 0.002 and 0.017], IDI: 0.284 and 0.201 [p = 0.005 and 0.043], respectively; in external test set 1, NRI: 0.183 and 0.161 [p = 0.019 and 0.008], IDI: 0.205 and 0.184 [p = 0.031 and 0.045], respectively; in external test set 2, NRI: 0.297 and 0.297 [p = 0.038 and 0.047], IDI: 0.332 and 0.294 [p = 0.031 and 0.041], respectively). The deep learning model constructed for automatic segmentation of ultrasound images can improve the diagnostic performance of radiologists for PGTs.

Limited capability of MRI radiomics to predict primary tumor histology of brain metastases in external validation.

Growing research demonstrates the ability to predict histology or genetic information of various malignancies using radiomic features extracted from imaging data. This study aimed to investigate MRI-based radiomics in predicting the primary tumor of brain metastases through internal and external validation, using oversampling techniques to address the class imbalance. This IRB-approved retrospective multicenter study included brain metastases from lung cancer, melanoma, breast cancer, colorectal cancer, and a combined heterogenous group of other primary entities (5-class classification). Local data were acquired between 2003 and 2021 from 231 patients (545 metastases). External validation was performed with 82 patients (280 metastases) and 258 patients (809 metastases) from the publicly available Stanford BrainMetShare and the University of California San Francisco Brain Metastases Stereotactic Radiosurgery datasets, respectively. Preprocessing included brain extraction, bias correction, coregistration, intensity normalization, and semi-manual binary tumor segmentation. Two-thousand five hundred and twenty-eight radiomic features were extracted from T1w (± contrast), fluid-attenuated inversion recovery (FLAIR), and wavelet transforms for each sequence (8 decompositions). Random forest classifiers were trained with selected features on original and oversampled data (5-fold cross-validation) and evaluated on internal/external holdout test sets using accuracy, precision, recall, F1 score, and area under the receiver-operating characteristic curve (AUC). Oversampling did not improve the overall unsatisfactory performance on the internal and external test sets. Incorrect data partitioning (oversampling before train/validation/test split) leads to a massive overestimation of model performance. Radiomics models' capability to predict histologic or genomic data from imaging should be critically assessed; external validation is essential.

Prediction of High-Risk Neuroblastoma Among Neuroblastic Tumors Using Radiomics Features Derived from Magnetic Resonance Imaging: A Pilot Study.

This study aimed to predict high-risk neuroblastoma among neuroblastic tumors using radiomics features extracted from MRI. Pediatric patients (age≤18 years) diagnosed with neuroblastic tumors who had pre-treatment MR images available were enrolled from institution A from January 2010 to November 2019 (training set) and institution B from January 2016 to January 2022 (test set). Segmentation was performed with regions of interest manually drawn along tumor margins on the slice with the widest tumor area by two radiologists. First-order and texture features were extracted and intraclass correlation coefficients (ICCs) were calculated. Multivariate logistic regression (MLR) and random forest (RF) models from 10-fold cross-validation were built using these features. The trained MLR and RF models were tested in an external test set. Thirty-two patients (M:F=23:9, 26.0±26.7 months) were in the training set and 14 patients (M:F=10:4, 33.4±20.4 months) were in the test set with radiomics features (n=930) being extracted. For 10 of the most relevant features selected, intra- and inter-observer variability was moderate to excellent (ICCs 0.633-0.911, 0.695-0.985, respectively). The area under the receiver operating characteristic curve (AUC) was 0.94 (sensitivity 67%, specificity 91%, and accuracy 84%) for the MLR model and the average AUC was 0.83 (sensitivity 44%, specificity 87%, and accuracy 75%) for the RF model from 10-fold cross-validation. In the test set, AUCs of the MLR and RF models were 0.94 and 0.91, respectively. An MRI-based radiomics model can help predict high-risk neuroblastoma among neuroblastic tumors.

Comparing three-dimensional and two-dimensional deep-learning, radiomics, and fusion models for predicting occult lymph node metastasis in laryngeal squamous cell carcinoma based on CT imaging: a multicentre, retrospective, diagnostic study

The occult lymph node metastasis (LNM) of laryngeal squamous cell carcinoma (LSCC) affects the treatment and prognosis of patients. This study aimed to comprehensively compare the performance of the three-dimensional and two-dimensional deep learning models, radiomics model, and the fusion models for predicting occult LNM in LSCC. In this retrospective diagnostic study, a total of 553 patients with clinical N0 stage LSCC, who underwent surgical treatment without distant metastasis and multiple primary cancers, were consecutively enrolled from four Chinese medical centres between January 01, 2016 and December 30, 2020. The participant data were manually retrieved from medical records, imaging databases, and pathology reports. The study cohort was divided into a training set (n=300), an internal test set (n=89), and two external test sets (n=120 and 44, respectively). The three-dimensional deep learning (3D DL), two-dimensional deep learning (2D DL), and radiomics model were developed using CT images of the primary tumor. The clinical model was constructed based on clinical and radiological features. Two fusion strategies were utilized to develop the fusion model: the feature-based DLRad_FB model and the decision-based DLRad_DB model. The discriminative ability and correlation of 3D DL, 2D DL and radiomics features were analysed comprehensively. The performances of the predictive models were evaluated based on the pathological diagnosis. The 3D DL features had superior discriminative ability and lower internal redundancy compared to 2D DL and radiomics features. The DLRad_DB model achieved the highest AUC (0.89-0.90) among all the study sets, significantly outperforming the clinical model (AUC=0.73-0.78, P=0.0001-0.042, Delong test). Compared to the DLRad_DB model, the AUC values for the DLRad_FB, 3D DL, 2D DL, and radiomics models were 0.82-0.84 (P=0.025-0.46), 0.86-0.89 (P=0.75-0.97), 0.83-0.86 (P=0.029-0.66), and 0.79-0.82 (P=0.0072-0.10), respectively in the study sets. Additionally, the DLRad_DB model exhibited the best sensitivity (82-88%) and specificity (79-85%) in the test sets. The decision-based fusion model DLRad_DB, which combines 3D DL, 2D DL, radiomics, and clinical data, can be utilized to predict occult LNM in LSCC. This has the potential to minimize unnecessary lymph node dissection and prophylactic radiotherapy in patients with cN0 disease. National Natural Science Foundation of China, Natural Science Foundation of Shandong Province.

Development and Multicenter, Multiprotocol Validation of Neural Network for Aberrant Right Subclavian Artery Detection.

This study aimed to develop and validate a convolutional neural network (CNN) that automatically detects an aberrant right subclavian artery (ARSA) on preoperative computed tomography (CT) for thyroid cancer evaluation. A total of 556 CT with ARSA and 312 CT with normal aortic arch from one institution were used as the training set for model development. A deep learning model for the classification of patch images for ARSA was developed using two-dimension CNN from EfficientNet. The diagnostic performance of our model was evaluated using external test sets (112 and 126 CT) from two institutions. The performance of the model was compared with that of radiologists for detecting ARSA using an independent dataset of 1683 consecutive neck CT. The performance of the model was achieved using two external datasets with an area under the curve of 0.97 and 0.99, and accuracy of 97% and 99%, respectively. In the temporal validation set, which included a total of 20 patients with ARSA and 1663 patients without ARSA, radiologists overlooked 13 ARSA cases. In contrast, the CNN model successfully detected all the 20 patients with ARSA. We developed a CNN-based deep learning model that detects ARSA using CT. Our model showed high performance in the multicenter validation.

Quality assurance of late gadolinium enhancement cardiac magnetic resonance images: a deep learning classifier for confidence in the presence or absence of abnormality with potential to prompt real-time image optimization

BackgroundLate gadolinium enhancement (LGE) of the myocardium has significant diagnostic and prognostic implications, with even small areas of enhancement being important. Distinguishing between definitely normal and definitely abnormal LGE images is usually straightforward, but diagnostic uncertainty arises when reporters are not sure whether the observed LGE is genuine or not. This uncertainty might be resolved by repetition (to remove artifact) or further acquisition of intersecting images, but this must take place before the scan finishes. Real-time quality assurance by humans is a complex task requiring training and experience, so being able to identify which images have an intermediate likelihood of LGE while the scan is ongoing, without the presence of an expert is of high value. This decision-support could prompt immediate image optimization or acquisition of supplementary images to confirm or refute the presence of genuine LGE. This could reduce ambiguity in reports. MethodsShort-axis, phase-sensitive inversion recovery late gadolinium images were extracted from our clinical cardiac magnetic resonance (CMR) database and shuffled. Two, independent, blinded experts scored each individual slice for “LGE likelihood” on a visual analog scale, from 0 (absolute certainty of no LGE) to 100 (absolute certainty of LGE), with 50 representing clinical equipoise. The scored images were split into two classes—either “high certainty” of whether LGE was present or not, or “low certainty.” The dataset was split into training, validation, and test sets (70:15:15). A deep learning binary classifier based on the EfficientNetV2 convolutional neural network architecture was trained to distinguish between these categories. Classifier performance on the test set was evaluated by calculating the accuracy, precision, recall, F1-score, and area under the receiver operating characteristics curve (ROC AUC). Performance was also evaluated on an external test set of images from a different center. ResultsOne thousand six hundred and forty-five images (from 272 patients) were labeled and split at the patient level into training (1151 images), validation (247 images), and test (247 images) sets for the deep learning binary classifier. Of these, 1208 images were “high certainty” (255 for LGE, 953 for no LGE), and 437 were “low certainty”. An external test comprising 247 images from 41 patients from another center was also employed. After 100 epochs, the performance on the internal test set was accuracy = 0.94, recall = 0.80, precision = 0.97, F1-score = 0.87, and ROC AUC = 0.94. The classifier also performed robustly on the external test set (accuracy = 0.91, recall = 0.73, precision = 0.93, F1-score = 0.82, and ROC AUC = 0.91). These results were benchmarked against a reference inter-expert accuracy of 0.86. ConclusionDeep learning shows potential to automate quality control of late gadolinium imaging in CMR. The ability to identify short-axis images with intermediate LGE likelihood in real-time may serve as a useful decision-support tool. This approach has the potential to guide immediate further imaging while the patient is still in the scanner, thereby reducing the frequency of recalls and inconclusive reports due to diagnostic indecision.

Preoperative CT-based intra- and peri-tumoral radiomic models for differentiating benign and malignant tumors of the parotid gland: a two-center study.

To investigate the ability of intra- and peritumoral radiomics based on three-phase computed tomography (CT) to distinguish between malignant and benign parotid tumors. We conducted a retrospective analysis of data from 374 patients with parotid gland tumors, all confirmed by histopathology. A total of 321 patients from Center 1 (January 2014 to January 2023) were randomly divided into the training set and internal testing set at a ratio of 7:3, whereas 53 patients from Center 2 (January 2020 to June 2022) constituted the external testing set. CT images of both the tumor and surrounding areas (2 mm and 5 mm areas surrounding the tumor) were reviewed, and their radiomic features were extracted for the construction of different radiomic models. In addition, a combined clinical-radiomic model was developed using multivariate logistic regression analysis. The model's predictive performance was evaluated using decision curve analysis (DCA) and receiver operating characteristic (ROC) curves. Among the models evaluated, Tumor + External2 model demonstrated superior predictive performance. The areas under the curve (AUCs) of this model were 0.986 in the training set, 0.827 in the internal test set, and 0.749 in the external test set. For the clinical model, independent predictive factors included symptoms, boundaries, and lymph node swelling. The combined clinical-radiomic model achieved AUCs of 0.981, 0.842, and 0.749 in the three cohorts, outperforming both the Tumor model and the clinical model individually. The CT-based radiomic models incorporating intratumoral and peritumoral radiomic features can effectively distinguish malignant from benign parotid tumors, and the predictive accuracy is further improved by incorporating clinically independent predictors.

Improved detection of aortic dissection in non-contrast-enhanced chest CT using an attention-based deep learning model

Rationale and objectivesThis study investigated the effects of implementing an attention-based deep learning model for the detection of aortic dissection (AD) using non-contrast-enhanced chest computed tomography (CT). Materials and methodsWe analysed the records of 1300 patients who underwent contrast-enhanced chest CT at 2 medical centres between January 2015 and February 2023. We considered an internal cohort of 200 patients with AD and 200 patients without AD and an external test cohort of 40 patients with AD and 40 patients without AD. The internal cohort was divided into training and test sets, and a deep learning model was trained using 9600 CT images. A convolutional block attention module (CBAM) and a traditional deep learning architecture (namely, You Only Look Once version 5 [YOLOv5]) were combined into an attention-based model (i.e., YOLOv5-CBAM). Its performance was measured against the unmodified YOLOv5 model, and the accuracy, sensitivity, and specificity of the algorithm were evaluated by two independent radiologists. ResultsThe CBAM-based model outperformed the traditional deep learning model. In the external testing set, YOLOv5-CBAM achieved an area under the curve (AUC) of 0.938, accuracy of 91.5 %, sensitivity of 90.0 %, and specificity of 92.9 %, whereas the unmodified model achieved an AUC of 0.844, accuracy of 83.6 %, sensitivity of 71.2 %, and specificity of 96.0 %. The sensitivity results of the unmodified algorithms were not significantly different from those of the radiologists; however, the proposed YOLOv5-CBAM algorithm outperformed the unmodified algorithms in terms of detection. ConclusionsIncorporating the CBAM attention mechanism into a deep learning model can significantly improve AD detection in non-contrast-enhanced chest CT. This approach may aid radiologists in the timely and accurate diagnosis of AD, which is important for improving patient outcomes.

Prediction of lateral lymph node metastasis in rectal cancer patients based on MRI using clinical, deep transfer learning, radiomic, and fusion models.

Lateral lymph node (LLN) metastasis in rectal cancer significantly affects patient treatment and prognosis. This study aimed to comprehensively compare the performance of various predictive models in predicting LLN metastasis. In this retrospective study, data from 152 rectal cancer patients who underwent lateral lymph node (LLN) dissection were collected. The cohort was divided into a training set (n=86) from Tianjin Union Medical Center (TUMC), and two testing cohorts: testing cohort (TUMC) (n=37) and testing cohort from Gansu Provincial Hospital (GSPH) (n=29). A clinical model was established using clinical data; deep transfer learning models and radiomics models were developed using MRI images of the primary tumor (PT) and largest short-axis LLN (LLLN), visible LLN (VLLN) areas, along with a fusion model that integrates features from both deep transfer learning and radiomics. The diagnostic value of these models for LLN metastasis was analyzed based on postoperative LLN pathology. Models based on LLLN image information generally outperformed those based on PT image information. Rradiomics models based on LLLN demonstrated improved robustness on external testing cohorts compared to those based on VLLN. Specifically, the radiomics model based on LLLN imaging achieved an AUC of 0.741 in the testing cohort (TUMC) and 0.713 in the testing cohort (GSPH) with the extra trees algorithm. Data from LLLN is a more reliable basis for predicting LLN metastasis in rectal cancer patients with suspicious LLN metastasis than data from PT. Among models performing adequately on the internal test set, all showed declines on the external test set, with LLLN_Rad_Models being less affected by scanning parameters and data sources.

Application of PLS-NN model based on mid-infrared spectroscopy in the origin identification of Cornus officinalis.

Mid-infrared spectroscopy has been increasingly used as a nondestructive analytical technique in Chinese herbal medicine identification in recent years. In this study, a new chemometric model named as PLS-NN model was proposed based on the mid-infrared spectral data of Cornus officinalis samples from 11 origins. It was realized by combining the partial least squares and neural networks for the identification of the origin of Chinese herbal medicines. First, we extracted features from the spectral data in 3448 bands using the partial least squares method, and extracted 122 components that contained more than 95% of the information. Then, we trained the PLS-NN model by neural network using the extracted components as inputs and the corresponding origin classes as outputs. Finally, based on an external test set, we evaluated the generalization ability of the PLS-NN model using metrics such as accuracy, F1-Score and Kappa coefficient. The results show that the PLS-NN model performs well in all three metrics when compared to models such as Decision trees, Support vector machine, Partial least squares Discriminant analysis, and Naive bayes. The model not only realizes the dimensionality reduction of full-spectrum data and improves the training efficiency of the model, but also has higher accuracy compared with the full-spectrum data model. The PLS-NN model was applied to identify the origin of Cornus officinalis with an accuracy of 91.9%.