Maximum Intensity Projection Research Articles

MEDS-Net: Multi-encoder based self-distilled network with bidirectional maximum intensity projections fusion for lung nodule detection

Early detection of lung cancer increases cure rates, making precise detection of lung nodules in computed tomography (CT) images crucial. However, creating an accurate computer-aided detection (CADe) system is challenging due to lung structure complexity and nodule heterogeneity. We present a lung nodule detection scheme, inspired by radiologist workflow, fusing CT scan sub-volumes with bidirectional maximum intensity projection (MIP) images within a single architecture. We have developed a novel multi-encoder-based network (MEDS-Net) using self-distillation to learn from three types of inputs: 3D sub-volume, forward, and backward MIP images. MEDS-Net utilizes three encoders to extract insights, passed to the decoder via attention units to suppress redundant features. The decoder employs a self-distillation mechanism, connecting to a distillation block with four auxiliary lung nodule detectors, accelerating convergence and enhancing learning ability. The auxiliary detectors are used during inference to reduce false positives. Our scheme, tested on 888 LIDC-IDRI dataset scans, achieved a 93.6% competition performance metric score, illustrating that fusing bidirectional MIP images with raw CT slices and auxiliary detectors empowers MEDS-Net to accurately detect lung nodules while minimizing false positives.

Improved small vessel visibility in diabetic foot arteriography using dual-energy CT

To test the feasibility and performance of dual-energy computed tomography (DECT) in foot arteriography of diabetic patients, where contrast medium is largely reduced within the small vessels. A total of 50 diabetic patients were enrolled prospectively, where DECT was acquired immediately after the CT angiography (CTA, group A) of the lower extremity. Two images were derived from the DECT data, one optimal virtual monochromatic image (VMI, group B) and one fusion image (group C), both of which were compared against the CTA image for visualising the foot arteries. The contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) were evaluated. The arterial course and contrast were graded each using a five-point scale. The clarity of small vessel depiction was quantified by comparing the number of plantar metatarsal arteries found in the maximum intensity projection image. The median CNRs and SNRs obtained in group B were approximately 45% and 20% higher than those in groups A and C, respectively (p<0.05). Group B also received higher subjective scores on the posterior tibial artery and the foot arteries (all >3) than groups A and C. The number of visible branches of the plantar metatarsal arteries was found to be substantially higher (p<0.05) in group B (median=6) than in groups A (median=2) and C (median=4). DECT was found to be superior to conventional CTA in foot arteriography, and beyond the lower extremity, it might be a general favourable solution for imaging regions with small vessels and reduced contrast medium.

DeAPIR: Efficient and Enhanced Dental Arch-Guided Panoramic Image Reconstruction from CBCT Data

Dental CBCT and panoramic imaging play a pivotal role in dental diagnosis and treatment planning, alongside the indispensable use of computed tomography (CT) and X-ray imaging in dentistry, particularly for surgical planning. Given the widespread adoption of dental cone beam CT in clinics today, we explore a novel approach in this study—utilizing CT’s three-dimensional (3D) data to reconstruct a two-dimensional (2D) panoramic dental image. This method eliminates the requirement for an extra panoramic scan. In this work, we propose a novel framework to generate an enhanced and extended 2D panoramic view by using the dental arch extracted from 3D CBCT. Our method involves segmenting the patient’s dental arch from their 3D CBCT image by identifying horizontal slices with above-average intensity, followed by morphological operations, including dilation, Gaussian filtering, and skeletonization, to delineate the dental arch line. Additionally, we extend the dental arch beyond the wisdom teeth using quadratic curve fitting. Finally, we employ Maximum Intensity Projection on rotated cubic segments aligned with the dental arch curve to produce captivating panoramic images. The panoramic view produced using our proposed method, when compared to the results obtained from BlueSky and OpenInventor, exhibited superior enhancements and greater accuracy in panoramic visualization.

Artificial intelligence for the detection of sacroiliitis on magnetic resonance imaging in patients with axial spondyloarthritis.

Magnetic resonance imaging (MRI) is important for the early detection of axial spondyloarthritis (axSpA). We developed an artificial intelligence (AI) model for detecting sacroiliitis in patients with axSpA using MRI. This study included MRI examinations of patients who underwent semi-coronal MRI scans of the sacroiliac joints owing to chronic back pain with short tau inversion recovery (STIR) sequences between January 2010 and December 2021. Sacroiliitis was defined as a positive MRI finding according to the ASAS classification criteria for axSpA. We developed a two-stage framework. First, the Faster R-CNN network extracted regions of interest (ROIs) to localize the sacroiliac joints. Maximum intensity projection (MIP) of three consecutive slices was used to mimic the reading of two adjacent slices. Second, the VGG-19 network determined the presence of sacroiliitis in localized ROIs. We augmented the positive dataset six-fold. The sacroiliitis classification performance was measured using the sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). The prediction models were evaluated using three-round three-fold cross-validation. A total of 296 participants with 4,746 MRI slices were included in the study. Sacroiliitis was identified in 864 MRI slices of 119 participants. The mean sensitivity, specificity, and AUROC for the detection of sacroiliitis were 0.725 (95% CI, 0.705-0.745), 0.936 (95% CI, 0.924-0.947), and 0.830 (95%CI, 0.792-0.868), respectively, at the image level and 0.947 (95% CI, 0.912-0.982), 0.691 (95% CI, 0.603-0.779), and 0.816 (95% CI, 0.776-0.856), respectively, at the patient level. In the original model, without using MIP and dataset augmentation, the mean sensitivity, specificity, and AUROC were 0.517 (95% CI, 0.493-0.780), 0.944 (95% CI, 0.933-0.955), and 0.731 (95% CI, 0.681-0.780), respectively, at the image level and 0.806 (95% CI, 0.729-0.883), 0.617 (95% CI, 0.523-0.711), and 0.711 (95% CI, 0.660-0.763), respectively, at the patient level. The performance was improved by MIP techniques and data augmentation. An AI model was developed for the detection of sacroiliitis using MRI, compatible with the ASAS criteria for axSpA, with the potential to aid MRI application in a wider clinical setting.

Assessment of left atrial fibrosis by 18F-FDG PET/CT in persistent atrial fibrillation : comparison to electroanatomical mapping and cMR late gadolinium enhancement

Abstract A wide body of evidence suggests that left atrial (LA) interstitial fibrosis is a key contributor to persistent atrial fibrillation (peAF) and that its extent is predictive of the success of catheter ablation (CA) for this condition. Late gadolinium enhanced magnetic resonance imaging (LGE-cMR) has become increasingly popular for non-invasive detection and quantification of atrial fibrosis prior to CA. However, it is limited by a high rate of suboptimal scans and by large variability of quantification of fibrosis. We recently described the ability to obtain high quality 18- fluorodeoxyglucose (FDG) positron emission tomography (PET) of the atria after acipimox administration, and hypothesize that, similar to the left ventricle (LV), atrial metabolic activity could be a marker of cellular viability, indicating the absence of atrial fibrosis. The objective of our study was to investigate the potential utility of 18FDG uptake in the LA as an indicator of atrial scar and a predictor of CA success in patients with peAF. To this end, we compared the LA 18FDG uptake to LA voltage mapping during electrophysiological studies (EP) and cMR-LGE as a reference standard. Method 33 nondiabetic patients with nonvalvular peAF hospitalized for a first CA in our center prospectively underwent 18FDG-PET/CT, and LGE-CMR prior to EP and CA procedures. FDG-PET, LGE-cMR and electoanatomical maps were segmented and corregistrated in 3D surface models based on anatomical information from the contrast enhanced cMR angiograms. For 18FDG-PET/CT, maximum intensity projection of standardized uptake values (SUVs) along the surface was calculated and abnormal low metabolism was identified using a threshold of four standard deviations below the average SUV value of the atrial blood. Electrical scarring on EP voltage maps was defined as bipolar measurements of &amp;lt; 0.3 mV in AF, while fibrosis by LGE-CMR was evaluated using several thresholds of 2-5 standard deviations above the average blood signal intensity and the average intensity of the enhancement around the mitral valve. All three parameters of scar were expressed in percentage of the total LA area. Results Atrial hypometabolism by FDG-PET correlated strongly with electrical scaring by low voltage area during EP study (p&amp;lt;0.001, R² = 0.573). Atrial wall mean SUV standardized by LV uptake was also associated with low voltage, also the association's strength was mild (p =0.012, R²=0.22). There was good spatial corregistration of areas of low FDG metabolism with low voltage areas during EP study (Figure 2). By contrast there was no association between PET and LGE-cMR and association between atrial LGE by cMR and atrial voltage for all used thresholds. Conclusion Low 18FDG uptake by PET/CT showed a good correlation with atrial electrical scaring, while LGE-CMR did not. Our data suggest that atrial FDG PET could therefore be a valuable technique for non-invasive evaluation of atrial fibrosis prior to CA.Low metabolism versus low voltageMetabolic and voltage maps

Dosimetric comparison of free-breathing versus respiratory motion-managed radiotherapy via four-dimensional computed tomography-based volumetric-modulated arctherapy for lung cancer

PurposeThe aim of this study is to use respiratory motion-managed radiotherapy (RT) to reduce side effects and to compare dosimetric factors with free-breathing planning in patients with lung cancer. Materials and methodsSimulation images were obtained in 10 respiratory phases with free breathing using four-dimensional computed tomography (4D-CT) scanner. Planning target volume (PTV) was created with 5mm margins in each direction of the internal target volume delineated using the maximum intensity projection. A volumetric arc treatment (VMAT) plan was created so that the prescribed dose would cover 98% of the PTV. Target volumes for the free-breathing VMAT plan were created according to ICRU Reports 62 and the same prescribed dose was used. ResultsPatients were evaluated during January 2020. Median 63Gy (59.4–64) RT was administered. Median PTV volumes were 173.53 and 494.50cm3 (P=0.008) and dose covering 95% of PTV volume was 62.97 and 60.51Gy (P=0.13) in 4D-CT based and free-breathing VMAT plans, respectively. The mean and V50 heart dose was 6.03Gy (vs. 10.36Gy, P=0.043) and 8.2% (vs. 33.9%, P=0.007), and significantly lower in 4D-CT based VMAT plans and there was also found a non-significant reduction for other risky organ doses. ConclusionTen patients treated with respiratory motion-managed RT with 4D-CT based VMAT technique. It was observed that PTV did not increase, the target was covered with 95% accuracy, and with statistical significance in heart doses, all risky organ doses were found to be less.

“Transcallosal” periventricular anastomosis in moyamoya disease: the fourth periventricular anastomosis and a potential predictor of hemorrhage

ObjectivesChoroidal anastomosis is a risk factor for hemorrhage in moyamoya disease. One variant of choroidal anastomosis, “transcallosal anastomosis,” originates from the medial posterior choroidal artery, and penetrates the corpus callosum to reconstruct the pericallosal artery. We aimed to investigate the prevalence and the bleeding rate of transcallosal anastomosis using sliding thin-slab maximum intensity projection reformatted from magnetic resonance angiography (MRA). Materials and MethodsThis study included 222 patients. We defined transcallosal anastomosis grades (0-2) and the stenosis of the anterior (ACA, 0-2), middle (MCA, 1-3), and posterior cerebral artery (PCA, 0-2) by MRA scores, independently by two coauthors. ResultsGrade-2 transcallosal anastomosis was detected in 21 patients (9.5 %). There were no correlations of the incidence of transcallosal anastomosis with previous bypass surgery (P = 0.23). Multivariate analysis revealed a significantly higher incidence in hemorrhagic onset and younger age (odds ratio [OR] 3.77, and 0.97). Transcallosal anastomosis had statistically significant correlation with ACA and PCA scores (P = 0.01 and 0.03), but not with MCA scores (P = 0.1). In multivariate analysis, ACA scores 1 and 2 were significantly higher (OR, 15.44 and 11.17), and PCA score 1 was also higher (OR, 3.07), but PCA score 2 was not. Interrater agreement for judgment of transcallosal anastomosis grade was strong (κ = 0.89). Two patients with Grade-2 transcallosal anastomosis had late hemorrhage in the corpus callosum (bleeding rate: 2.5 % per year). ConclusionsTranscallosal anastomosis may be associated with both advanced ACA and moderate PCA stenosis, and cause hemorrhage at the corpus callosum.

Maximum intensity projection based on high frame rate contrast-enhanced ultrasound for the differentiation of breast tumors.

We explored the role of maximum intensity projection (MIP) based on high frame rate contrast-enhanced ultrasound (H-CEUS) for the differentiation of breast tumors. MIP imaging was performed in patients with breast tumors who underwent H-CEUS examinations. The microvasculature morphology of breast tumors was assessed. The receiver operating characteristic curve was plotted to evaluate the diagnostic performance of MIP. Forty-three breast tumors were finally analyzed, consisting of 19 benign and 24 malignant tumors. For the ≤30-s and >30-s phases, dot-, line-, or branch-like patterns were significantly more common in benign tumors. A tree-like pattern was only present in the benign tumors. A crab claw-like pattern was significantly more common in the malignant tumors. Among the tumors with crab claw-like patterns, three cases of malignant tumors had multiple parallel small spiculated vessels. There were significant differences in the microvasculature morphology for the ≤30-s and >30-s phases between the benign and malignant tumors (all p < 0.001). The area under the curve, sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of the ≤30-s phase were all higher than those of the >30-s phase for the classification of breast tumors. MIP based on H-CEUS can be used for the differentiation of breast tumors, and the ≤30-s phase had a better diagnostic value. Multiple parallel small spiculated vessels were a new finding, which could provide new insight for the subsequent study of breast tumors.

Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules.

For patients with early multiple pulmonary nodules, it is essential, from a diagnostic perspective, to determine the spatial distribution, size, location, and relationship with surrounding lung tissue of these nodules throughout the entire lung. This is crucial for identifying the primary lesion and developing more scientifically grounded treatment plans for doctors. However, pattern recognition methods based on machine vision are susceptible to false positives and false negatives and, therefore, cannot fully meet clinical demands in this regard. Visualization methods based on maximum intensity projection (MIP) can better illustrate local and individual pulmonary nodules but lack a macroscopic and holistic description of the distribution and spatial features of multiple pulmonary nodules. Therefore, this study proposes a whole-lung 3D reconstruction method. It extracts the 3D contour of the lung using medical image processing technology against the background of the entire lung and performs 3D reconstruction of the lung, pulmonary artery, and multiple pulmonary nodules in 3D space. This method can comprehensively depict the spatial distribution and radiological features of multiple nodules throughout the entire lung, providing a simple and convenient means of evaluating the diagnosis and prognosis of multiple pulmonary nodules.

Facilitating cell segmentation with the projection-enhancement network

Contemporary approaches to instance segmentation in cell science use 2D or 3D convolutional networks depending on the experiment and data structures. However, limitations in microscopy systems or efforts to prevent phototoxicity commonly require recording sub-optimally sampled data that greatly reduces the utility of such 3D data, especially in crowded sample space with significant axial overlap between objects. In such regimes, 2D segmentations are both more reliable for cell morphology and easier to annotate. In this work, we propose the projection enhancement network (PEN), a novel convolutional module which processes the sub-sampled 3D data and produces a 2D RGB semantic compression, and is trained in conjunction with an instance segmentation network of choice to produce 2D segmentations. Our approach combines augmentation to increase cell density using a low-density cell image dataset to train PEN, and curated datasets to evaluate PEN. We show that with PEN, the learned semantic representation in CellPose encodes depth and greatly improves segmentation performance in comparison to maximum intensity projection images as input, but does not similarly aid segmentation in region-based networks like Mask-RCNN. Finally, we dissect the segmentation strength against cell density of PEN with CellPose on disseminated cells from side-by-side spheroids. We present PEN as a data-driven solution to form compressed representations of 3D data that improve 2D segmentations from instance segmentation networks.

Computed tomography and magnetic resonance imaging of diffuse axonal injury in brain trauma in young children: retrospective study

INTRODUCTION: Traumatic brain injury (TBI) is the leading cause of death and disability in children. Children’s TBI is associated with a number of characteristics that distinguish it from adults. Although the death rate associated with TBI has decreased over the past 2 decades, the disability of children who survived TBI continues to have a significant impact on the economy and public health of society as a whole.OBJECTIVE: To show the possibilities of computed tomography and magnetic resonance imaging in the diagnosis of traumatic diffuse axonal injuries in infants and young children.MATERIALS AND METHODS: In this work, we analyzed the CT and MRI data of 1334 children under 3 years of age with acute TBI who were treated at the Clinical and Research Institute of Emergency Pediatric Surgery and Trauma (730 boys, 604 girls). The age of the children was from 23 days to 3 years, the average age was 1 year 6 months. Computed tomography scans were performed on a 128-slice Ingenuity Elite scanner (Philips). Scanning of the area of interest (head + cervical spine) was carried out with the maximum possible reduction in indicators to minimize the radiation dose, including the iDose4 program. MRI was performed on a Phillips Achieva 3 T scanner with multiplanar T1- and T2WI, 2D and 3D isotropic images, FLAIR, SWI, DWI/DTI, and magnetic resonance angiography (MRA). No contrast enhancement was applied. Statistics: data processing was carried out using the GraphPad Prism 9 software package. The computing and graphic capabilities of the Excel spreadsheet editor were used. To compare the mean values of the data samples, ANOVA analysis of variance wasused; differences were considered significant at a significance level of p&lt;0.05.RESULTS: 824 (61.8%) of 1334 children had a concussion, 510 (38.2%) had traumatic injuries of skull and brain from uncomplicated cephalohematomas and linear fractures to massive intracranial hematomas and total brain edema. Diffuse axonal injuries(DAI) type I and II on CT, further confirmed by MRI, were detected in 32 (6.27%) of all 510 children with TBI. 19 out of 32 hada combination with brain contusions, 13 with epi — and / or subdural hematomas of a small volume. MRI was performed in 89 out of 510 children under 3 years of age with intracranial injuries, in whom CT was initially performed. DAI of various types were detected by MRI in 92.13% (82/89) of the patients.DISCUSSION: DAI is one of the most common types of TBI, occurring in both mild and severe forms, and is a brain injury characterized by axonal disruption, resulting in lesions of white matter tracts over a wide area. Taking into account the fact that children with TBI were admitted to the institute with a significant spread (from 1 hour to 7 days), the initial CT scan was performed no later than one hour after admission and was supplemented by MRI. The criteria for CT and MRI prescribements after TBI were: clinical and neurological status at the time of admission. CT has low sensitivity for detecting DAI, as only large hemorrhagic (hyperdense) lesions of DAI are detected. The use of multiplanar reconstruction, as well as various options, such as 3D reconstruction, MIP and MinIP (maximum and minimum intensity projection) and others, allows for a detailed assessment of the brain parenchyma and increases the detection of DAI. In our study, DAI using MRI was detected in 92.13% of patients, and the most critical type of DAI III was detected in 100% of children with severe TBI with poor outcome and death of patients.CONCLUSION: Head CT is the preferred method for emergency radiology of TBI patients due to its affordability, speed of data acquisition, and ability to detect lesions requiring urgent neurosurgical intervention. In acute cases, MRI is chosen for patients with severe neurological impairment despite the absence of structural brain damage on CT. MRI is the method of choice for subacute and chronic TBI

Application of Reciprocity for Facilitation of Wave Field Visualization and Defect Detection

ABSTRACT The motion visualization in a structural component was studied for defect detection. Elastic motions were excited by hammer impacts at multiple points and received by an accelerometer at a fixed point. Reciprocity in elastodynamics is only valid under certain conditions. Its validity under given experimental conditions was derived from the elastodynamic reciprocity theorem. Based on this, the dynamic motion of the structural component was obtained for fixed-point excitation from measurements performed using multipoint excitations. In the visualized eigenmodes, significant additional deformation was observed at the wall thinning inserted as an artificial defect. To prevent the dependence of defect detection on its position within the mode shape, another approach was proposed based on the extraction of guided wave modes immediately after impact excitation. It is shown that this maximum intensity projection method works well in detecting defects.

New Data-Driven Gated (DDG) PET/CT for Radiation Treatment Planning of NSCLC

PET/CT for RT simulation of NSCLC is typically conducted in two separate imaging sessions of gated 4D-CT in radiation oncology and fusion with non-gated PET/CT in diagnostic imaging, making treatment response assessment (TRA) with PET/CT and radiation treatment planning (RTP) challenging. To remedy this short-coming, we designed a new data-driven gated (DDG) PET/CT, based on the self-gated PET and self-gated CT data to derive the gated PET/CT data free of tumor blurring and misregistration artifacts without any respiratory monitoring device (ease of use and cost reduction). The new DDG PET/CT also provides the important 4D-CT information of end-inspiratory (EI), end-expiratory (EE), maximum-intensity-projection (MIP) and average-intensity-projection (AIP) CT's. The total imaging time is < 15 min, which makes DDG PET/CT simulation in a single session without any respiratory monitoring device possible. Acquisition is standard PET/CT imaging protocol followed by a cine CT imaging of the tumor area. We utilized the commercial DDG PET from GE Discovery PET/CT and designed a new DDG CT for attenuation correction of the DDG PET for reduction of tumor motion and mis-registration. The AIP and MIP images were from the average and maximum pixel values of the cine CT images, respectively. The cine CT images with the largest and smallest average CT number in the lung region were identified as the EE and EI CT images, respectively. For the image slices without any lung region present, the largest and smallest expansions of the body outline contour were selected to be the EE and EI CT phases, respectively. The AIP and EE CT images were used for attenuation correction of the PET and DDG PET data, respectively. Both DDG PET/CT and 4D-CT of (EI, EE, AIP and MIP) are for RT simulation of NSCLC. 38 4D-CT patient data sets were compared at the EE and EI phases between our DDG CT and 4D-CT to demonstrate the applicability of the newly designed DDG CT. A prototype software on a Dell PC of i5-6500 CPU has been successfully developed to enable DDG PET/CT and 4D-CT on seven GE Discovery PET/CT scanners. The operating system was Ubuntu and the computer language was Python. In the EE phase, the images selected by DDG CT and 4D CT were identical 62.5±21.6% of the time. In the EI phase, the images selected by DDG CT and 4D CT were identical 68.2±18.9% of the time. Inspection of the EE and EI phases of DDG CT and 50% and 0% of 4D-CT demonstrated both data sets of DDG CT and 4D CT were almost identical. A new DDG PET/CT and 4D-CT in a single imaging session without any hardware gating has been developed for TRA and RTP of NSCLC. The utility is available on the network and it enables all PET/CT scanners on the network for TRA and RTP. The new DDG CT provides the benefits of 4D-CT without any requirement of external hardware gating and make possible DDG PET/CT free of tumor motion and misregistration artifacts.

Sensitivity of an AI method for [18F]FDG PET/CT outcome prediction of diffuse large B-cell lymphoma patients to image reconstruction protocols

BackgroundConvolutional neural networks (CNNs), applied to baseline [18F]-FDG PET/CT maximum intensity projections (MIPs), show potential for treatment outcome prediction in diffuse large B-cell lymphoma (DLBCL). The aim of this study is to investigate the robustness of CNN predictions to different image reconstruction protocols. Baseline [18F]FDG PET/CT scans were collected from 20 DLBCL patients. EARL1, EARL2 and high-resolution (HR) protocols were applied per scan, generating three images with different image qualities. Image-based transformation was applied by blurring EARL2 and HR images to generate EARL1 compliant images using a Gaussian filter of 5 and 7 mm, respectively. MIPs were generated for each of the reconstructions, before and after image transformation. An in-house developed CNN predicted the probability of tumor progression within 2 years for each MIP. The difference in probabilities per patient was then calculated between both EARL2 and HR with respect to EARL1 (delta probabilities or ΔP). We compared these to the probabilities obtained after aligning the data with ComBat using the difference in median and interquartile range (IQR).ResultsCNN probabilities were found to be sensitive to different reconstruction protocols (EARL2 ΔP: median = 0.09, interquartile range (IQR) = [0.06, 0.10] and HR ΔP: median = 0.1, IQR = [0.08, 0.16]). Moreover, higher resolution images (EARL2 and HR) led to higher probability values. After image-based and ComBat transformation, an improved agreement of CNN probabilities among reconstructions was found for all patients. This agreement was slightly better after image-based transformation (transformed EARL2 ΔP: median = 0.022, IQR = [0.01, 0.02] and transformed HR ΔP: median = 0.029, IQR = [0.01, 0.03]).ConclusionOur CNN-based outcome predictions are affected by the applied reconstruction protocols, yet in a predictable manner. Image-based harmonization is a suitable approach to harmonize CNN predictions across image reconstruction protocols.

Convolutional neural network-based program to predict lymph node metastasis of non-small cell lung cancer using 18F-FDG PET.

To develop a convolutional neural network (CNN)-based program to analyze maximum intensity projection (MIP) images of 2-deoxy-2-[F-18]fluoro-D-glucose (FDG) positron emission tomography (PET) scans, aimed at predicting lymph node metastasis of non-small cell lung cancer (NSCLC), and to evaluate its effectiveness in providing diagnostic assistance to radiologists. We obtained PET images of NSCLC from public datasets, including those of 435 patients with available N-stage information, which were divided into a training set (n = 304) and a test set (n = 131). We generated 36 maximum intensity projection (MIP) images for each patient. A residual network (ResNet-50)-based CNN was trained using the MIP images of the training set to predict lymph node metastasis. Lymph node metastasis in the test set was predicted by the trained CNN as well as by seven radiologists twice: first without and second with CNN assistance. Diagnostic performance metrics, including accuracy and prediction error (the difference between the truth and the predictions), were calculated, and reading times were recorded. In the test set, 67 (51%) patients exhibited lymph node metastases and the CNN yielded 0.748 predictive accuracy. With the assistance of the CNN, the prediction error was significantly reduced for six of the seven radiologists although the accuracy did not change significantly. The prediction time was significantly reduced for five of the seven radiologists with the median reduction ratio 38.0%. The CNN-based program could potentially assist radiologists in predicting lymph node metastasis by increasing diagnostic confidence and reducing reading time without affecting diagnostic accuracy, at least in the limited situations using MIP images.

Deep Learning Reconstruction Improves the Image Quality of CT Angiography Derived From 80-kVp Cerebral CT Perfusion Data

To investigate the impact of the deep learning reconstruction (DLR) technique on the image quality of CT angiography (CTA) derived from 80-kVp cerebral CT perfusion (CTP) data and compare it with hybrid-iterative reconstruction (HIR). Thirty-three patients underwent CTP at 80 kVp were prospectively enrolled. CTP data were reconstructed with HIR and DLR. Four image datasets were reconstructed: HIRpeak and DLRpeak were single arterial phase images derived from the time point showing the peak value, HIRtMIP and HIRtAve were time-resolved maximum intensity projection image and time-resolved average image derived from three time points with the greatest enhancement of HIR. The mean CT values, standard deviation, signal-to-noise ratio, and contrast-to-noise ratio of the internal carotid artery and basilar artery were compared among the four image dataset. Image quality was performed using a five-point rating scale. Arterial stenosis was evaluated. DLRpeak had the highest CT value and contrast-to-noise ratio in the internal carotid artery and basilar artery (all p < 0.001). DLRpeak showed the best subjective image quality and had the highest score (4.93 ± 0.4) compared to the other three HIR CTA images (all p < 0.001). The degree of vascular stenosis was consistent among the four evaluated sequences (HIRtAve, HIRpeak, and HIRtMIP DLRpeak). For CTA derived from 80-kVp cerebral CTP data, images reconstructed with deep learning showed better image quality and improved intracranial artery visualization than those processed with HIR and other currently used techniques.

Prevalence and influencing factors for artifact development in breast MRI-derived maximum intensity projections.

Magnetic resonance imaging (MRI) provides high diagnostic sensitivity for breast cancer. However, MRI artifacts may impede the diagnostic assessment. This is particularly important when evaluating maximum intensity projections (MIPs), such as in abbreviated MRI (AB-MRI) protocols, because high image quality is desired as a result of fewer sequences being available to compensate for problems. To describe the prevalence of artifacts on dynamic contrast enhanced (DCE) MRI-derived MIPs and to investigate potentially associated attributes. For this institutional review board approved retrospective analysis, MIPs were generated from subtraction series and cropped to represent the left and right breasts as regions of interest. These images were labeled by three independent raters regarding the presence of MRI artifacts. MRI artifact prevalence and associations with patient characteristics and technical attributes were analyzed using descriptive statistics and generalized linear models (GLMMs). The study included 2524 examinations from 1794 patients (median age 50 years), performed on 1.5 and 3.0 Tesla MRI systems. Overall inter-rater agreement was kappa = 0.54. Prevalence of significant unilateral artifacts was 29.2% (736/2524), whereas bilateral artifacts were present in 37.8% (953/2524) of all examinations. According to the GLMM, artifacts were significantly positive associated with age (odds ratio [OR]=1.52) and magnetic field strength (OR=1.55), whereas a negative effect could be shown for body mass index (OR=0.95). MRI artifacts on DCE subtraction MIPs of the breast, as used in AB-MRI, are a relevant topic. Our results show that, besides the magnetic field strength, further associated attributes are patient age and body mass index, which can provide possible targets for artifact reduction.

Localization of contrast-enhanced breast lesions in ultrafast screening MRI using deep convolutional neural networks

ObjectivesTo develop a deep learning–based method for contrast-enhanced breast lesion detection in ultrafast screening MRI.Materials and methodsA total of 837 breast MRI exams of 488 consecutive patients were included. Lesion’s location was independently annotated in the maximum intensity projection (MIP) image of the last time-resolved angiography with stochastic trajectories (TWIST) sequence for each individual breast, resulting in 265 lesions (190 benign, 75 malignant) in 163 breasts (133 women). YOLOv5 models were fine-tuned using training sets containing the same number of MIP images with and without lesions. A long short-term memory (LSTM) network was employed to help reduce false positive predictions. The integrated system was then evaluated on test sets containing enriched uninvolved breasts during cross-validation to mimic the performance in a screening scenario.ResultsIn five-fold cross-validation, the YOLOv5x model showed a sensitivity of 0.95, 0.97, 0.98, and 0.99, with 0.125, 0.25, 0.5, and 1 false positive per breast, respectively. The LSTM network reduced 15.5% of the false positive prediction from the YOLO model, and the positive predictive value was increased from 0.22 to 0.25.ConclusionsA fine-tuned YOLOv5x model can detect breast lesions on ultrafast MRI with high sensitivity in a screening population, and the output of the model could be further refined by an LSTM network to reduce the amount of false positive predictions.Clinical relevance statementThe proposed integrated system would make the ultrafast MRI screening process more effective by assisting radiologists in prioritizing suspicious examinations and supporting the diagnostic workup.Key Points• Deep convolutional neural networks could be utilized to automatically pinpoint breast lesions in screening MRI with high sensitivity.• False positive predictions significantly increased when the detection models were tested on highly unbalanced test sets with more normal scans.• Dynamic enhancement patterns of breast lesions during contrast inflow learned by the long short-term memory networks helped to reduce false positive predictions.

Imaging features of cardioembolic stroke on 4-dimensional computed tomography angiography.

Identifying cardioembolic stroke is important for the decision-making of endovascular treatment and anticoagulation therapy. We aimed to explore the features of cardioembolic stroke on 4-dimensional (4D) computed tomography angiography (4D-CTA) and assess whether these features can assist in classifying stroke etiology. In this retrospective study, we analyzed the images of 294 patients with acute ischemic stroke (AIS) from July 2020 to February 2022 at the First Affiliated Hospital of Chongqing Medical University, which had been consecutively collected. The data of 110 patients with occlusion of the M1/M2 segment of the middle cerebral artery (MCA) with/without intracranial internal carotid artery (ICA) occlusion were analyzed to calculate the clot burden score (CBS) and collateral score (CS), and the data of 88 patients with a clear origin and distal part were analyzed to measure clot length. Maximum intensity projection (MIP) and time MIP (tMIP) post-processing were used to assess the clot features. The Mann-Whitney U test was used to compare the clot characteristics between the 2 groups. Binary logistic regression was performed to assess the association between the image characteristics and cardioembolic stroke. Moreover, the receiver operating characteristic (ROC) curve was used to test the diagnostic efficacy of MIP/tMIP clot features in classifying cardioembolic stroke. Age, high-risk factors for cerebrovascular disease, high/medium-risk sources of cardioembolic stroke, clot length, CBS, and CS were significantly different between the cardioembolic stroke group and non-cardioembolic stroke group (P<0.05). In the cardioembolic stroke group, the median MIP and tMIP clot length was 12 mm [interquartile range (IQR), 8.3-17.4 mm] and 9.3 mm (IQR, 6.8-14.3 mm), respectively. In the non-cardioembolic stroke group, the median MIP and tMIP clot length was 6.5 mm (IQR, 4.7-11.5 mm) and 5.8 mm (IQR, 3.9-10.6 mm), respectively. Binary logistic regression showed that cardioembolic stroke was significantly associated with MIP-clot length [odds ratio (OR), 1.15; 95% confidence interval (CI): 1.02-1.29; P<0.05], tMIP-clot length (OR, 1.18; 95% CI: 1.02-1.36; P<0.05), and tMIP-CBS (OR, 3.96; 95% CI: 1.08-14.58; P<0.05). The area under the ROC curve (AUC) values of MIP clot length for identifying cardioembolic stroke were 0.75 (95% CI: 0.65-0.84, P<0.05), with a cut-off value of >7.4 mm [sensitivity: 84.62% (95% CI: 69.50-94.10%); specificity: 59.18% (95% CI: 44.20-73.00%)]. The AUC value of tMIP clot length was 0.72 (95% CI: 0.61-0.81, P<0.05), with a cut-off value of >5.4 mm [sensitivity: 92.31% (95% CI: 79.10-98.40%); specificity: 48.98% (95% CI: 34.40-63.70%)]. Clot length and CBS were overestimated on MIP images. Among the clot characteristics, clot length could identify cardioembolic stroke.

A simple mathematical method to identify optimal biplane fluoroscopic angulations for chronic total occlusion percutaneous coronary intervention using CT angiography.

The concept of three-dimensional (3D) wiring for chronic total occlusion (CTO) percutaneous coronary intervention (PCI) is now widely accepted among coronary interventionalists. The 3 axes, i.e., the 2 X-ray beams and the CTO segment, should intersect with each other at as close to a right angle as possible. However, how to specify optimal fluoroscopic angulations for a given CTO segment has not been well established. We aimed to develop a simple and practical method to identify optimal fluoroscopic angulations for CTO PCI. A CTO vector can be derived from slab maximum intensity projection (MIP) images of coronary computed tomography (CT) angiography. Using trigonometric functions, the inner product of vectors and the equation of a plane, we calculated 2 fluoroscopic vectors perpendicular to each other and to the CTO vector. We applied this method to a patient with mid-left circumflex CTO and translated the resulting fluoroscopic vectors into optimal fluoroscopic angulations. To facilitate its use, we developed a calculator using spreadsheet software that can output optimal fluoroscopic angulations within a practical range by inputting the x, y, and z components of the CTO vector. This approach also helps to minimise dead angles in biplane fluoroscopy. This method has the potential to make CTO PCI safer and easier, without requiring dedicated equipment or software. Its effectiveness should be validated in clinical practice.