Standalone deep learning versus experts for diagnosis lung cancer on chest computed tomography: a systematic review.

PURPOSE: To compare the diagnostic performance of standalone deep learning (DL) algorithms and human experts in lung cancer detection on chest computed tomography (CT) scans. MATERIALS AND METHODS: This study searched for studies on PubMed, Embase, and Web of Science from their inception until November 2023. We focused on adult lung cancer patients and compared the efficacy of DL algorithms and expert radiologists in disease diagnosis on CT scans. Quality assessment was performed using QUADAS-2, QUADAS-C, and CLAIM. Bivariate random-effects and subgroup analyses were performed for tasks (malignancy classification vs invasiveness classification), imaging modalities (CT vs low-dose CT [LDCT] vs high-resolution CT), study region, software used, and publication year. RESULTS: We included 20 studies on various aspects of lung cancer diagnosis on CT scans. Quantitatively, DL algorithms exhibited superior sensitivity (82%) and specificity (75%) compared to human experts (sensitivity 81%, specificity 69%). However, the difference in specificity was statistically significant, whereas the difference in sensitivity was not statistically significant. The DL algorithms' performance varied across different imaging modalities and tasks, demonstrating the need for tailored optimization of DL algorithms. Notably, DL algorithms matched experts in sensitivity on standard CT, surpassing them in specificity, but showed higher sensitivity with lower specificity on LDCT scans. CONCLUSION: DL algorithms demonstrated improved accuracy over human readers in malignancy and invasiveness classification on CT scans. However, their performance varies by imaging modality, underlining the importance of continued research to fully assess DL algorithms' diagnostic effectiveness in lung cancer. CLINICAL RELEVANCE STATEMENT: DL algorithms have the potential to refine lung cancer diagnosis on CT, matching human sensitivity and surpassing in specificity. These findings call for further DL optimization across imaging modalities, aiming to advance clinical diagnostics and patient outcomes. KEY POINTS: Lung cancer diagnosis by CT is challenging and can be improved with AI integration. DL shows higher accuracy in lung cancer detection on CT than human experts. Enhanced DL accuracy could lead to improved lung cancer diagnosis and outcomes.

Deep Learning Detection and Segmentation of Brain Arteriovenous Malformation on Magnetic Resonance Angiography.

BACKGROUND: The delineation of brain arteriovenous malformations (bAVMs) is crucial for subsequent treatment planning. Manual segmentation is time-consuming and labor-intensive. Applying deep learning to automatically detect and segment bAVM might help to improve clinical practice efficiency. PURPOSE: To develop an approach for detecting bAVM and segmenting its nidus on Time-of-flight magnetic resonance angiography using deep learning methods. STUDY TYPE: Retrospective. SUBJECTS: 221 bAVM patients aged 7-79 underwent radiosurgery from 2003 to 2020. They were split into 177 training, 22 validation, and 22 test data. FIELD STRENGTH/SEQUENCE: 1.5 T, Time-of-flight magnetic resonance angiography based on 3D gradient echo. ASSESSMENT: The YOLOv5 and YOLOv8 algorithms were utilized to detect bAVM lesions and the U-Net and U-Net++ models to segment the nidus from the bounding boxes. The mean average precision, F1, precision, and recall were used to assess the model performance on the bAVM detection. To evaluate the model's performance on nidus segmentation, the Dice coefficient and balanced average Hausdorff distance (rbAHD) were employed. STATISTICAL TESTS: The Student's t-test was used to test the cross-validation results (P < 0.05). The Wilcoxon rank test was applied to compare the median for the reference values and the model inference results (P < 0.05). RESULTS: The detection results demonstrated that the model with pretraining and augmentation performed optimally. The U-Net++ with random dilation mechanism resulted in higher Dice and lower rbAHD, compared to that without that mechanism, across varying dilated bounding box conditions (P < 0.05). When combining detection and segmentation, the Dice and rbAHD were statistically different from the references calculated using the detected bounding boxes (P < 0.05). For the detected lesions in the test dataset, it showed the highest Dice of 0.82 and the lowest rbAHD of 5.3%. DATA CONCLUSION: This study showed that pretraining and data augmentation improved YOLO detection performance. Properly limiting lesion ranges allows for adequate bAVM segmentation. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY STAGE: 1.

Federated Learning: A Cross-Institutional Feasibility Study of Deep Learning Based Intracranial Tumor Delineation Framework for Stereotactic Radiosurgery.

BACKGROUND: Deep learning-based segmentation algorithms usually required large or multi-institute data sets to improve the performance and ability of generalization. However, protecting patient privacy is a key concern in the multi-institutional studies when conventional centralized learning (CL) is used. PURPOSE: To explores the feasibility of a proposed lesion delineation for stereotactic radiosurgery (SRS) scheme for federated learning (FL), which can solve decentralization and privacy protection concerns. STUDY TYPE: Retrospective. SUBJECTS: 506 and 118 vestibular schwannoma patients aged 15-88 and 22-85 from two institutes, respectively; 1069 and 256 meningioma patients aged 12-91 and 23-85, respectively; 574 and 705 brain metastasis patients aged 26-92 and 28-89, respectively. FIELD STRENGTH/SEQUENCE: 1.5T, spin-echo, and gradient-echo [Correction added after first online publication on 21 August 2023. Field Strength has been changed to "1.5T" from "5T" in this sentence.]. ASSESSMENT: The proposed lesion delineation method was integrated into an FL framework, and CL models were established as the baseline. The effect of image standardization strategies was also explored. The dice coefficient was used to evaluate the segmentation between the predicted delineation and the ground truth, which was manual delineated by neurosurgeons and a neuroradiologist. STATISTICAL TESTS: The paired t-test was applied to compare the mean for the evaluated dice scores (p < 0.05). RESULTS: FL performed the comparable mean dice coefficient to CL for the testing set of Taipei Veterans General Hospital regardless of standardization and parameter; for the Taichung Veterans General Hospital data, CL significantly (p < 0.05) outperformed FL while using bi-parameter, but comparable results while using single-parameter. For the non-SRS data, FL achieved the comparable applicability to CL with mean dice 0.78 versus 0.78 (without standardization), and outperformed to the baseline models of two institutes. DATA CONCLUSION: The proposed lesion delineation successfully implemented into an FL framework. The FL models were applicable on SRS data of each participating institute, and the FL exhibited comparable mean dice coefficient to CL on non-SRS dataset. Standardization strategies would be recommended when FL is used. LEVEL OF EVIDENCE: 4 TECHNICAL EFFICACY: Stage 1.

Neurobiological Markers for Predicting Treatment Response in Patients with Bipolar Disorder.

Predictive neurobiological markers for prognosis are essential but underemphasized for patients with bipolar disorder (BD), a neuroprogressive disorder. Hence, we developed models for predicting symptom and functioning changes. Sixty-one patients with BD were recruited and assessed using the Young Mania Rating Scale (YMRS), Montgomery−Åsberg Depression Rating Scale (MADRS), Positive and Negative Syndrome Scale (PANSS), UKU Side Effect Rating Scale (UKU), Personal and Social Performance Scale (PSP), and Global Assessment of Functioning scale both at baseline and after 1-year follow-up. The models for predicting the changes in symptom and functioning scores were trained using data on the brain morphology, functional connectivity, and cytokines collected at baseline. The correlation between the predicted and actual changes in the YMRS, MADRS, PANSS, and UKU scores was higher than 0.86 (q < 0.05). Connections from subcortical and cerebellar regions were considered for predicting the changes in the YMRS, MADRS, and UKU scores. Moreover, connections of the motor network were considered for predicting the changes in the YMRS and MADRS scores. The neurobiological markers for predicting treatment-response symptoms and functioning changes were consistent with the neuropathology of BD and with the differences found between treatment responders and nonresponders.

Automated coronary artery calcium scoring using nested U-Net and focal loss.

Coronary artery calcium (CAC) is a great risk predictor of the atherosclerotic cardiovascular disease and CAC scores can be used to stratify the risk of heart disease. Current clinical analysis of CAC is performed using onsite semiautomated software. This semiautomated CAC analysis requires experienced radiologists and radiologic technologists and is both demanding and time-consuming. The purpose of this study is to develop a fully automated CAC detection model that can quantify CAC scores. A total of 1,811 cases of cardiac examinations involving contrast-free multidetector computed tomography were retrospectively collected. We divided the database into the Training Data Set, Validation Data Set, Testing Data Set 1, and Testing Data Set 2. The Training, Validation, and Testing Data Set 1 contained cases with clinically detected CAC; Testing Data Set 2 contained those without detected calcium. The intraclass correlation coefficients between the overall standard and model-predicted scores were 1.00 for both the Training Data Set and Testing Data Set 1. In Testing Data Set 2, the model was able to detect clinically undetected cases of mild calcium. The results suggested that the proposed model's automated detection of CAC was highly consistent with clinical semiautomated CAC analysis. The proposed model demonstrated potential for clinical applications that can improve the quality of CAC risk stratification.

Acceleration of Magnetic Resonance Fingerprinting Reconstruction Using Denoising and Self-Attention Pyramidal Convolutional Neural Network.

Magnetic resonance fingerprinting (MRF) based on echo-planar imaging (EPI) enables whole-brain imaging to rapidly obtain T1 and T2* relaxation time maps. Reconstructing parametric maps from the MRF scanned baselines by the inner-product method is computationally expensive. We aimed to accelerate the reconstruction of parametric maps for MRF-EPI by using a deep learning model. The proposed approach uses a two-stage model that first eliminates noise and then regresses the parametric maps. Parametric maps obtained by dictionary matching were used as a reference and compared with the prediction results of the two-stage model. MRF-EPI scans were collected from 32 subjects. The signal-to-noise ratio increased significantly after the noise removal by the denoising model. For prediction with scans in the testing dataset, the mean absolute percentage errors between the standard and the final two-stage model were 3.1%, 3.2%, and 1.9% for T1, and 2.6%, 2.3%, and 2.8% for T2* in gray matter, white matter, and lesion locations, respectively. Our proposed two-stage deep learning model can effectively remove noise and accurately reconstruct MRF-EPI parametric maps, increasing the speed of reconstruction and reducing the storage space required by dictionaries.

Evaluation of Abdominal Computed Tomography Scans for Differentiating the Discrepancies in Abdominal Adipose Tissue Between Two Major Subtypes of Primary Aldosteronism.

The aim of this study was to analyze the differences in the distribution of abdominal adipose tissue between the two subtypes of primary aldosteronism (PA) using abdominal computed tomography. We retrospectively analyzed patients diagnosed as having essential hypertension (EH) or PA from the prospectively collected Taiwan Primary Aldosteronism Investigation (TAIPAI) database. Patients with PA were divided into the subgroups of idiopathic hyperaldosteronism (IHA) and unilateral aldosterone-producing adenoma (APA). Patients' basic clinicodemographic data were collected, and a self-developed CT-based software program was used to quantify the abdominal adiposity indexes, including visceral adipose tissue (VAT) area, VAT ratio, waist circumference (WC), subcutaneous adipose tissue (SAT) area, and SAT ratio. We included 190 patients with EH and 436 patients with PA (238 with IHA and 198 with APA). The APA group had significantly lower abdominal adiposity indexes than the other groups. We also found negative correlations of aldosterone-to-renin ratio (ARR) with VAT area, VAT ratio, WC, and body mass index (BMI) in the APA group. After propensity score matching (which left 184 patients each in the IHA and APA groups), patients in the APA group still had significantly lower WC, SAT area, SAT ratio, and VAT ratio than those in the IHA group. Furthermore, logistic regression analysis indicated that lower probability of abdominal obesity was significantly related to patients with APA. Our data revealed that the distribution of abdominal adipose tissue was similar in patients with IHA and those with EH, but the abdominal adiposity indexes were significantly lower in patients with APA than in those with IHA and EH.

Aldosterone-producing adenoma-harbouring KCNJ5 mutations is associated with lower prevalence of metabolic disorders and abdominal obesity.

OBJECTIVE: Aldosterone overproduction and lipid metabolic disturbances between idiopathic hyperaldosteronism (IHA) and unilateral aldosterone-producing adenoma (APA) have been inconsistently linked in patients with primary aldosteronism. Moreover, KCNJ5 mutations are prevalent among APAs and enhance aldosterone synthesis in adrenal cortex. We aimed to investigate the prevalence of metabolic syndrome (MetS) in each primary aldosteronism subtype and observe the role of KCNJ5 mutations among APAs on the distribution of abdominal adipose tissues quantified using computed tomography (CT), including their changes postadrenalectomy. DESIGN AND METHODS: We retrospectively collected 244 and 177 patients with IHA and APA at baseline. Patients with APA had undergone adrenalectomy, and gene sequencing revealed the absence (n = 75) and presence (n = 102) of KCNJ5 mutations. We also recruited 31 patients with APA who had undergone CT-scan 1-year postadrenalectomy. RESULTS: The patients with APA harbouring KCNJ5 mutations had significantly lower prevalence of MetS and smaller distribution in waist circumference, subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT) than the other groups. Logistic regression analysis indicated that the VAT area correlated significantly with KCNJ5 mutations among the APAs. Only participants with KCNJ5 mutations had significant increases in triglycerides, cholesterol, SAT, and VAT after 1-year postadrenalectomy. CONCLUSION: This study is the first to demonstrate that MetS and abdominal obesity were less prevalent in the patients with APA harbouring KCNJ5 mutations compared with the IHA group and the non-KCNJ5-mutated APA group. Increasing prevalence of dyslipidaemia and abdominal obesity was observed in patients with KCNJ5 mutations 1-year postadrenalectomy.

Accelerated white matter lesion analysis based on simultaneous T1 and T2∗ quantification using magnetic resonance fingerprinting and deep learning.

PURPOSE: To develop an accelerated postprocessing pipeline for reproducible and efficient assessment of white matter lesions using quantitative magnetic resonance fingerprinting (MRF) and deep learning. METHODS: MRF using echo-planar imaging (EPI) scans with varying repetition and echo times were acquired for whole brain quantification of T1 and T2∗ in 50 subjects with multiple sclerosis (MS) and 10 healthy volunteers along 2 centers. MRF T1 and T2∗ parametric maps were distortion corrected and denoised. A CNN was trained to reconstruct the T1 and T2∗ parametric maps, and the WM and GM probability maps. RESULTS: Deep learning-based postprocessing reduced reconstruction and image processing times from hours to a few seconds while maintaining high accuracy, reliability, and precision. Mean absolute error performed the best for T1 (deviations 5.6%) and the logarithmic hyperbolic cosinus loss the best for T2∗ (deviations 6.0%). CONCLUSIONS: MRF is a fast and robust tool for quantitative T1 and T2∗ mapping. Its long reconstruction and several postprocessing steps can be facilitated and accelerated using deep learning.

In-room assessment of intravascular velocity from time-resolved rotational angiography in patients with arteriovenous malformation: a pilot study.

BACKGROUND: Time-resolved rotational angiography (t-RA) enables interventionists to better comprehend complex arteriovenous malformations (AVMs), thereby facilitating endovascular treatment. However, its use in evaluating hemodynamic changes has rarely been explored. OBJECTIVE: This study uses t-RA to estimate intravascular flow in patients with AVM to compare this with flow in the normal population. METHODS: Patients with available t-RA scans were prospectively categorized into one of three groups: hemorrhagic AVM, non-hemorrhagic AVM and control. Pulsatile time-density curves (TDCs) for C1, C6 and VOIMCA were used for amplitude and velocity estimation. C1 was at the cervical internal carotid artery (ICA), 2-3 cm below the carotid canal, C6 was at the paraclinoid segment of the ICA, and VOIMCA was at the junction of the first and second segment of the middle cerebral artery (MCA). A waveform amplitude ratio was defined as (peak - trough)/trough contrast intensity. VICA was defined as the distance between C6 and C1 divided by the time required for the wave to pass, and correspondingly, the average velocity of MCA (VMCA) was defined as the distance between C6 and VOIMCA divided by the duration for the same peak to travel from C6 and VOIMCA, AVM volume was estimated by MR angiography. RESULTS: Amplitude ratios AC1 and AC6, and average flow velocities VICA and VMCA were significantly larger in the non-hemorrhagic group than in the control group, while the hemorrhagic AVM group was not significantly different from the controls. VICA and VMCA showed moderate to good correlations with AVM volume (r=0.51 and 0.73, respectively). VMCA (33.0±9.1) was significantly lower than VICA (41.3±13.2) in the control group, but not in the two AVM groups. CONCLUSION: TDC waveform propagation derived from t-RA can quantify hemodynamic differences between AVM and the control group. t-RA provides both real-time anatomic and hemodynamic evaluation, and can thus potentially improve the interventional workflow.

Automatic flow analysis of digital subtraction angiography using independent component analysis in patients with carotid stenosis.

PURPOSE: Current time-density curve analysis of digital subtraction angiography (DSA) provides intravascular flow information but requires manual vasculature selection. We developed an angiographic marker that represents cerebral perfusion by using automatic independent component analysis. MATERIALS AND METHODS: We retrospectively analyzed the data of 44 patients with unilateral carotid stenosis higher than 70% according to North American Symptomatic Carotid Endarterectomy Trial criteria. For all patients, magnetic resonance perfusion (MRP) was performed one day before DSA. Fixed contrast injection protocols and DSA acquisition parameters were used before stenting. The cerebral circulation time (CCT) was defined as the difference in the time to peak between the parietal vein and cavernous internal carotid artery in a lateral angiogram. Both anterior-posterior and lateral DSA views were processed using independent component analysis, and the capillary angiogram was extracted automatically. The full width at half maximum of the time-density curve in the capillary phase in the anterior-posterior and lateral DSA views was defined as the angiographic mean transient time (aMTT; i.e., aMTTAP and aMTTLat). The correlations between the degree of stenosis, CCT, aMTTAP and aMTTLat, and MRP parameters were evaluated. RESULTS: The degree of stenosis showed no correlation with CCT, aMTTAP, aMTTLat, or any MRP parameter. CCT showed a strong correlation with aMTTAP (r = 0.67) and aMTTLat (r = 0.72). Among the MRP parameters, CCT showed only a moderate correlation with MTT (r = 0.67) and Tmax (r = 0.40). aMTTAP showed a moderate correlation with Tmax (r = 0.42) and a strong correlation with MTT (r = 0.77). aMTTLat also showed similar correlations with Tmax (r = 0.59) and MTT (r = 0.73). CONCLUSION: Apart from vascular anatomy, aMTT estimates brain parenchyma hemodynamics from DSA and is concordant with MRP. This process is completely automatic and provides immediate measurement of quantitative peritherapeutic brain parenchyma changes during stenting.