Lung Nodule Texture Detection and Classification Using 3D CNN

Functional magnetic resonance imaging (fMRI) is one of the best techniques for precise localization of abnormal brain activity non-invasively. Machine-learning approaches have been widely used in neuroimaging studies; however, few studies have investigated the single-voxel modeling of fMRI data under cognitive tasks. We proposed a hybrid one-dimensional (1D) convolutional neural network (1D-CNN) based on the temporal dynamics of single-voxel fMRI time-series and successfully differentiated two continuous task states, namely, self-initiated (SI) and visually guided (VG) motor tasks. First, 25 activation peaks were identified from the contrast maps of SI and VG tasks in a blocked design. Then, the fMRI time-series of each peak voxel was transformed into a temporal-frequency domain by using continuous wavelet transform across a broader frequency range (0.003–0.313 Hz, with a step of 0.01 Hz). The transformed time-series was inputted into a 1D-CNN model for the binary classification of SI and VG continuous tasks. Compared with the univariate analysis, e.g., amplitude of low-frequency fluctuation (ALFF) at each frequency band, including, wavelet-ALFF, the 1D-CNN model highly outperformed wavelet-ALFF, with more efficient decoding models [46% of 800 models showing area under the curve (AUC) > 0.61] and higher decoding accuracies (94% of the efficient models), especially on the high-frequency bands (>0.1 Hz). Moreover, our results also demonstrated the advantages of wavelet decompositions over the original fMRI series by showing higher decoding performance on all peak voxels. Overall, this study suggests a great potential of single-voxel analysis using 1D-CNN and wavelet transformation of fMRI series with continuous, naturalistic, steady-state task design or resting-state design. It opens new avenues to precise localization of abnormal brain activity and fMRI-guided precision brain stimulation therapy.

Pancreatic cancer (PC) is frequently missed on unenhanced CT examinations performed for unrelated clinical indications, where the pancreas is included incidentally and clinical suspicion is low. To develop and validate a deep learning-based tool for PC diagnosis and risk stratification on unenhanced CT. This retrospective study included 3080 unenhanced CT studies of Taiwanese patients with PC, other pancreatic diseases and normal pancreas between 2004 and 2019 from a tertiary hospital, randomly divided into training, validation, and internal test sets. Unenhanced CT studies from United States institutions were used for external testing. A hybrid convolutional neural network-transformer model was trained for PC diagnosis and risk stratification. Performance was evaluated using sensitivity, specificity, and area under the curve (AUC), with comparisons to 2 radiologists by McNemar's test and exploratory decision curve analysis. The internal dataset included 713 PCs (mean age, 64.6 ± 12.0 years; 384 men), 1661 normal pancreas and 706 other pancreatic diseases. In an exploratory comparison restricted to unenhanced CT (29 PCs, 31 controls), the sensitivity of computer-aided diagnosis (CAD) tool (89.7%, 72.6-97.8) seemed comparable with that of 1 radiologist (86.2%, 68.3-96.1) and higher than another (41.4%, 23.5-61.1); but wide confidence intervals and inter-radiologist variability warrant cautious interpretation. In the internal test set (142 PCs, 474 controls), sensitivity was 90.8% (84.9-95.0) and specificity 93.0% (90.4-95.2) (AUC: 0.98), with sensitivity comparable to radiologist reports based on enhanced and unenhanced CT (95.4%, 90.2-98.3; P = .21). In the external set (42 PCs, 22 controls), sensitivity was 76.2% (60.5-87.9) and specificity 86.4% (65.1-97.1) (AUC: 0.89). The tool stratified cases into 7 risk levels with likelihood ratios ranging from <0.01 to 173.46. Exploratory decision curve analysis suggested potential net benefit across threshold probabilities. This tool may assist in the opportunistic detection and risk stratification of PC on unenhanced CT.

The advent of three-dimensional convolutional neural networks (3D CNNs) has revolutionized the detection and analysis of COVID-19 cases. As imaging technologies have advanced, 3D CNNs have emerged as a powerful tool for segmenting and classifying COVID-19 in medical images. These networks have demonstrated both high accuracy and rapid detection capabilities, making them crucial for effective COVID-19 diagnostics. This study offers a thorough review of various 3D CNN algorithms, evaluating their efficacy in segmenting and classifying COVID-19 across a range of medical imaging modalities. This review systematically examines recent advancements in 3D CNN methodologies. The process involved a comprehensive screening of abstracts and titles to ensure relevance, followed by a meticulous selection and analysis of research papers from academic repositories. The study evaluates these papers based on specific criteria and provides detailed insights into the network architectures and algorithms used for COVID-19 detection. The review reveals significant trends in the use of 3D CNNs for COVID-19 segmentation and classification. It highlights key findings, including the diverse range of networks employed for COVID-19 detection compared to other diseases, which predominantly utilize encoder/decoder frameworks. The study provides an in-depth analysis of these methods, discussing their strengths, limitations, and potential areas for future research. The study reviewed a total of 60 papers published across various repositories, including Springer and Elsevier. The insights from this study have implications for clinical diagnosis and treatment strategies. Despite some limitations, the accuracy and efficiency of 3D CNN algorithms underscore their potential for advancing medical image segmentation and classification. The findings suggest that 3D CNNs could significantly enhance the detection and management of COVID-19, contributing to improved healthcare outcomes.

Computer Aided Diagnosis (CAD) tools for classification of benign and malignant breast lesions are often applied and used as a “second opinion” to reduce misclassification rate. Most CAD tools rely rigidly on extraction of discriminatory feature sets. However, radiologist’s reasoning about mammographic meaning is often based on complex visual and non-visual characteristic aspects that are difficult to express in suitable discriminatory feature based representations. This is due to the fact that the interpretation of imagery data by a human involves various cognitive processes that are too complex to formulate in an algorithmic way. Consequently, in applying CAD solutions it is important for radiologists to realize how the CAD tool can negatively alter their visual cognitive reasoning. To address the above concern, we have developed an approach that seeks to aid radiologists by providing indexing and retrieval schemas based on contents-based information addressing rather than just the “second opinion” CAD tool. In this approach, a trained hybrid classifier is used to return a gallery of “similar” cases, represented as regions of interest (ROI), with their ground truth, i.e., ROI annotation, full mammogram, and/or biopsy results. By examining these similar cases and using recognition heuristics the radiologist makes his/her final determination.

Background: The computer-aided diagnosis (CAD) has been applied in multiple studies conducted recently in order to perform ultrasound (US) of the breast. There are several studies that have indicated that CAD is useful for improving the diagnostic performance in less experienced radiologists. However, there is no study on several readers analyzing the same lesions using a breast phantom and rare reliability studies. Objectives: To investigate the reliability of different readers using computer aided diagnosis (CAD) system in US for determining identical lesions in a breast phantom. Patients and Methods: From March 2016 to February 2017, six readers (three senior and three junior residents in the department of radiology) evaluated and analyzed breast phantom including 14 lesions. At the first line study, three senior residents (3rd grade with more than one month of training for breast ultrasound [US]) and three junior residents (1st grade without breast US training) evaluated and analyzed the US and applied CAD for lesions in breast phantom, and they were able to make final decisions by subjective combination. A month later, they conducted the second line study as they did the first line study. We analyzed the inter- and intra-reader reliability and accuracy of US, CAD, and combinations (subjective, conjunctive, and disjunctive). Results: In the total of first and second line studies of six readers, the kappa value of US (0.609) was significantly higher than CAD (0.411). In the subjective combined conclusion, the kappa value was improved by the junior group. In the whole inter- and intra-reader analysis, the kappa values of final assessment of the senior and junior group were more variable on CAD than on US, especially in the junior group, and this result was statistically significant. The area under the curve (AUC) of US, CAD, subjective, conjunctive, and disjunctive combination in seniors were all better than those of juniors. In all groups, the AUCs, sensitivities, and specificities were improved on conjunctive combined US with CAD. Conclusion: The combination US with CAD improved the reliability and diagnostic performance. The CAD results of the junior group were variable and inconsistent. Therefore, minimum training and experience for breast US is indispensable for the better use of breast CAD, and, combination US with CAD is useful for all readers.

Classification of brain tumors is of great importance in medical applications that benefit from computer-aided diagnosis. Misdiagnosis of brain tumor type will both prevent the patient from responding effectively to the applied treatment and decrease the patient’s chances of survival. In this study, we propose a solution for classifying brain tumors in MR images using transfer learning networks. The most common brain tumors are detected with VGG16, VGG19, ResNet50 and DenseNet21 networks using transfer learning. Deep transfer learning networks are trained and tested using four different optimization algorithms (Adadelta, ADAM, RMSprop and SGD) on the accessible Figshare dataset containing 3064 T1-weighted MR images from 233 patients with three common brain tumor types: glioma (1426 images), meningioma (708 images) and pituitary (930 images). The area under the curve (AUC) and accuracy metrics were used as performance measures. The proposed transfer learning methods have a level of success that can be compared with studies in the literature; the highest classification performance is 99.02% with ResNet50 using Adadelta. The classification result proved that the most common brain tumors can be classified with very high performance. Thus, the transfer learning model is promising in medicine and can help doctors make quick and accurate decisions.

Identification of faulty feeder in resonant grounding distribution networks remains a significant challenge due to the weak fault current and complicated working conditions. In this paper, we present a deep learning-based multi-label classification framework to reliably distinguish the faulty feeder. Three different neural networks (NNs) including the multilayer perceptron, one-dimensional convolutional neural network (1D CNN), and 2D CNN are built. However, the labeled data may be difficult to obtain in the actual environment. We use the simplified simulation model based on a full-scale test field (FSTF) to obtain sufficient labeled source data. Different from most learning-based methods assuming that the distribution of source domain and target domain is identical, we propose a samples-based transfer learning method to improve the domain adaptation by using samples in the source domain with appreciate weights. The TrAdaBoost algorithm is adopted to update the weights of each sample. The recorded data obtained in the FSTF are utilized to test the domain adaptability. According to our validation and testing, the validation accuracies are high when there is sufficient labeled data for training the proposed NNs. The proposed 2D CNN has the best domain adaptability. The TrAdaBoost algorithm can help the NNs to train an efficient classifier that has better domain adaptation. It has been therefore concluded that the proposed method especially the 2D CNN is suitable for actual distribution networks.

IntroductionsThe spine is one of the most common sites of metastasis for malignancies. This study aimed to compare the predictive performance of seven commonly used prognostic scoring systems for surgically treated spine metastases. It is expected to assist surgeons in selecting appropriate scoring systems to support clinical decision‐making and better inform patients.MethodsWe performed a retrospective study involving 268 surgically treated patients with spine metastases between 2017 and 2020 at a single regional oncology center in China. The revised Tokuhashi, Tomita, modified Bauer, revised Katagiri, van der Linden, Skeletal Oncology Research Group (SORG) nomogram, and SORG machine‐learning (ML) scoring systems were externally validated. The area under the curve (AUC) of the receiver operating characteristic curve was used to evaluate sensitivity and specificity at different postoperative time points. The actual survival time was compared with the reference survival time provided in the original publication.ResultsIn the present study, the median survival was 16.6 months. The SORG ML scoring system demonstrated the highest accuracy in predicting 90‐day (AUC: 0.743) and 1‐year survival (AUC: 0.787). The revised Katagiri demonstrated the highest accuracy (AUC: 0.761) in predicting 180‐day survival. The revised Katagiri demonstrated the highest accuracy (AUC: 0.779) in predicting 2‐year survival. Based on this series, the actual life expectancy was underestimated compared with the original reference survival time.ConclusionsNone of the scoring systems can perform optimally at all time points and for all pathology types, and the reference survival times provided in the original study need to be updated. A cautious awareness of the underestimation by these models is of paramount importance in relation to current patients.

ObjectivesTo compare the predictive ability of six pre-endoscopic scoring systems (ABC, AIMS65, GBS, MAP(ASH), pRS, and T-score) for outcomes of upper gastrointestinal bleeding (UGIB) in elderly and younger patients.MethodsA retrospective study of 1260 patients, including 530 elderly patients (age ge 65) and 730 younger patients (age < 65) presenting with UGIB, was performed at Zhongda Hospital Southeast University, from January 2015 to December 2020. Six scoring systems were used.ResultsABC had the largest areas under the curve (AUCs) of 0.827 (0.792–0.858), and 0.958 (0.929–0.987) for elderly and younger groups for predicting mortality respectively. The differences of the AUCs for predicting the outcome of mortality and rebleeding between the two groups were significant for ABC and pRS (p < 0.01). For intervention prediction, significant differences were observed only for pRS [AUC 0.623 (0.578–0.669) vs. 0.699 (0.646–0.752)] (p < 0.05) between the two groups. For intensive care unit (ICU) admission, the AUC for MAP (ASH) [0.791 (0.718–0.865) vs. 0.891 (0.831–0.950)] and pRS [0.610 (0.514–0.706) vs. 0.891 (0.699–0.865)] were more effective for the younger group (p < 0.05 and p < 0.01, respectively). For comparison of scoring systems in the same cohort, ABC was significantly higher than pRS: AUC 0.710 (0.699–0.853, p < 0.05) and T-score 0.670 (0.628–0.710, p < 0.01) for predicting mortality in the elderly group. In the younger group, ABC was significantly higher than GBS and T-score (p < 0.01). MAP(ASH) performs the best in predicting intervention in both groups.ConclusionsABC and pRS are more accurate for predicting mortality and rebleeding in the younger cohort, and pRS may not be suitable for elderly patients. There was no difference between the two study populations for GBS, AIMS65, and T-score. Except for ICU admission, MAP(ASH) showed fair accuracy for both cohorts.

Development and comparison of prognostic scoring systems for surgical closure of genitourinary fistula

BackgroundTo introduce a three-dimensional convolutional neural network (3D CNN) leveraging transfer learning for fusing PET/CT images and clinical data to predict EGFR mutation status in lung adenocarcinoma (LADC).MethodsRetrospective data from 516 LADC patients, encompassing preoperative PET/CT images, clinical information, and EGFR mutation status, were divided into training (n = 404) and test sets (n = 112). Several deep learning models were developed utilizing transfer learning, involving CT-only and PET-only models. A dual-stream model fusing PET and CT and a three-stream transfer learning model (TS_TL) integrating clinical data were also developed. Image preprocessing includes semi-automatic segmentation, resampling, and image cropping. Considering the impact of class imbalance, the performance of the model was evaluated using ROC curves and AUC values.ResultsTS_TL model demonstrated promising performance in predicting the EGFR mutation status, with an AUC of 0.883 (95%CI = 0.849–0.917) in the training set and 0.730 (95%CI = 0.629–0.830) in the independent test set. Particularly in advanced LADC, the model achieved an AUC of 0.871 (95%CI = 0.823–0.919) in the training set and 0.760 (95%CI = 0.638–0.881) in the test set. The model identified distinct activation areas in solid or subsolid lesions associated with wild and mutant types. Additionally, the patterns captured by the model were significantly altered by effective tyrosine kinase inhibitors treatment, leading to notable changes in predicted mutation probabilities.ConclusionPET/CT deep learning model can act as a tool for predicting EGFR mutation in LADC. Additionally, it offers clinicians insights for treatment decisions through evaluations both before and after treatment.

Objective To explore the predictive value of enhanced CURB score in the prognosis of patients with community-acquired pneumonia (CAP) in the setting of immunosuppressive therapy. Methods Retrospective study of 156 CAP patients with diseases treated with immunosuppressive agents admitted from January 2012 to July 2016 was carried out. The patients were divided into survival group and death group, and comparisons of demographics of patients and clinical sitting between two groups were performed. The receiver operator characteristic(ROC) curve was used for the calculations of enhanced CURB score, CURB-65 score and qSOFA score measured to predict the 28-day outcome of patients as the clinical observation endpoint. The predictive value of three scoring systems was compared in the prognosis of CAP patients under immunosuppressive treatment using the area under the curve (AUC). Results Of 156 patients, there were 134 patients in survival group and 22 patients in death group. The statistically significant differences in measured results of three scoring systems between two groups were as follows: enhanced CURB score [8(6, 10) vs. 12(9, 13)], CURB-65 score [1(0, 2) vs. 2(2, 3)], and sSOFA score [0(0, 1) vs. 1(0, 1)] (P<0.05 in 3 scoring systems). According to ROC, the AUC of enhanced CURB score was 0.815, with sensitivity 50%, specificity 97.01%, the optimal cut-off value 11 (P<0.01); the AUC of CURB-65 score was 0.816, with sensitivity 81.82%, specificity 65.67%, the best cut-off value 1 (P<0.01); the AUC of qSOFA score was 0.642, with sensitivity 54.5%, specificity 73.1%, the best cut-off value 0 (P<0.01). There was a significant difference in score between qSOFA score and CURB-65 score, between qSOFA score and enhanced CURB score, but no significant difference between enhanced CURB score and CURB-65 score. Conclusions The enhanced CURB score has preferable predictive value in evaluating the prognosis of CAP patients during immunosuppressive treatment. Though its sensitivity is low, its specificity is superior to CURB-65 score and qSOFA score. Key words: Community-acquired pneumonia; Immunosuppression; CURB-65; Enhanced CURB; qSOFA

Hepatic cystic echinococcosis (HCE), a life-threatening liver disease, has 5 subtypes, i.e., single-cystic, polycystic, internal capsule collapse, solid mass, and calcified subtypes. And each subtype has different treatment methods. An accurate diagnosis is the prerequisite for effective HCE treatment. However, clinicians with less diagnostic experience often make misdiagnoses of HCE and confuse its 5 subtypes in clinical practice. Computer-aided diagnosis (CAD) techniques can help clinicians to improve their diagnostic performance. This paper aims to propose an efficient CAD system that automatically differentiates 5 subtypes of HCE from the ultrasound images. The proposed CAD system adopts the concept of deep transfer learning and uses a pre-trained convolutional neural network (CNN) named VGG19 to extract deep CNN features from the ultrasound images. The proven classifier models, k - nearest neighbor (KNN) and support vecter machine (SVM) models, are integrated to classify the extracted deep CNN features. 3 distinct experiments with the same deep CNN features but different classifier models (softmax, KNN, SVM) are performed. The experiments followed 10 runs of the five-fold cross-validation process on a total of 1820 ultrasound images and the results were compared using Wilcoxon signed-rank test. The overall classification accuracy from low to high was 90.46 ± 1.59% for KNN classifier, 90.92 ± 2.49% for transfer learned VGG19, and 92.01 ± 1.48% for SVM, indicating SVM classifiers with deep CNN features achieved the best performance (P < 0.05). Other performance measures used in the study are specificity, sensitivity, precision, F1-score, and area under the curve (AUC). In addition, the paper addresses a practical aspect by evaluating the system with smaller training data to demonstrate the capability of the proposed classification system. The observations of the study imply that transfer learning is a useful technique when the availability of medical images is limited. The proposed classification system by using deep CNN features and SVM classifier is potentially helpful for clinicians to improve their HCE diagnostic performance in clinical practice.

e16572 Background: Presently, prostate biopsy (PBx) results typically report the highest Gleason Grade Group in the PBx as the single metric used to gauge the clinical aggressiveness of tumor and dictate treatment. However, using that single parameter alone limits the clinician by lack of consideration of the entire PBx. Intuitively we presumed that a PBx showing multiple cores of cancer would represent more aggressive disease. Herein, we propose the Weighted Gleason Grade Group (WGGG), a novel scoring system that synthesizes histopathologic data and cancer volume into a single numeric value representing the entire PBx, allowing the urologist to more accurately predict adverse pathologic outcomes of radical prostatectomy (RP). Methods: We studied 246 men who underwent RP after standard multi-core PBx. The highest Gleason score of each PBx was converted to its final GGG. The WGGG was calculated by summing the Gleason score of each positive core and normalized for a 12 core total. RP pathology was studied to determine adverse pathological outcomes, specifically, extraprostatic extension (EPE), positive surgical margins (PSM), seminal vesical invasion (SVI), pathological up-grading and any adverse feature. We then studied the ability of conventional GGG versus WGGG to have ‘predicted’ the risk of these adverse features comparing their respective receiver operating characteristic (ROC) areas under the curve (AUC) for each, as well as any adverse feature. Results: The AUC for the WGGG vs. GGG was significantly higher for predicting EPE (AUC 0.782 vs. 0.697, respectively; z = -3.29; p = .0009), PSM (AUC 0.644 vs. 0.563, respectively; z = -3.00; p = .0027), SVI (AUC 0.829 vs. 0.713, respectively; z = -3.05; p = .0023) and pathologic up-grading (AUC 0.584 vs. 0.349, respectively; z = -3.13; p = .002). Finally, the AUC for any adverse feature was 0.637 for WGGG versus 0.556 for GGG; z = -3.29; p = .001. Conclusions: The WGGG, by providing a metric reflecting the entirety of the PBx, is more informative than conventional single GGG alone in ‘predicting’ adverse pathologic outcomes on radical prostatectomy specimens. We are now studying if % cancer/ PBx core will improve WGGG performance.

Objective: Mortality prediction methods are still controversial about coronavirus disease-2019 pneumonia.This study aimed to compare the efficacy of the the quick Sequential Organ Failure Assessment, systemic inflammatory response syndrome (SIRS), Modified Early Warning score (MEWS), National Early Warning score, 4C mortality, and COVID-GRAM critical illness risk score (COVID-GRAM), scoring systems in predicting 28-day mortality in adult patients with COVID-19.Methods: This single-center, retrospective, observational cohort study included patients presenting to a pandemic hospital between November 2021 and December 2021.Inclusion criteria were patients aged 18 years or older, patients with positive reverse transcription-polymerase chain reaction test, and thoracic computed tomography imaging.The receiver operating characteristic analysis was performed to examine the diagnostic accuracy of the investigated scoring systems in predicting 28-day mortality.Statistical analyses were performed using the SPSS and MedCalc software packages.A p-value of <0.5 was considered statistically significant. Results:The study was conducted in 846 patients.The median age of the patients included in the study was 49 (36-75) years, and the rate of male patients was 46.3% (n=392).The rate of pneumonia detection was 85.1% (n=720).The hospitalization rate was 49.6% (n=420), the admission rate to the intensive care unit was 7.4% (n=63), and the 28-day mortality rate was 5.7% (n=48).The highest area under the curve (AUC) values for 28-day mortality prediction was obtained from COVID-GRAM (AUC: 0.935) and 4C mortality (AUC: 0.922) scores, while the lowest AUC values were calculated in SIRS (AUC: 0.756) and MEWS (AUC: 0.805). Conclusion:According to our results, COVID-GRAM may be the first-choice scoring system in the emergency department for predicting the 28-day mortality associated with COVID-19.