The Journal of Computer Assisted Tomography

04/06/2024

Current Issue: Effects of the Training Data Condition on Arterial Spin Labeling Parameter Estimation Using a Simulation-Based Supervised Deep Neural Network: Objective

A simulation-based supervised deep neural network (DNN) can accurately estimate cerebral blood flow (CBF) and arterial transit time (ATT) from multidelay arterial spin labeling signals. However, the performance of deep learning depends on the characteristics of the training data set. We aimed to investigate the effects of the ground truth (GT) ranges of CBF and ATT on the performance of the DNN when training data were prepared using arterial spin labeling signal simulation.

Methods

Deep neural networks were individually trained using 36 patterns of the training data sets. Simulation test data (1,000,000 points), 17 healthy volunteers, and 1 patient with moyamoya disease were included. The simulation test data were used to evaluate accuracy, precision, and noise immunity of the DNN. The best-performing DNN was determined by the normalized mean absolute error (NMAE), normalized root mean squared error (NRMSE), and normalized coefficient of variation over repeated training (CVNet). Cerebral blood flow and ATT values and their histograms were compared between the GT and predicted values. For the in vivo data, the dependency of the predicted values on the GT ranges was visually evaluated by comparing CBF and ATT maps between the best-performing DNN and the other DNNs. Moreover, using the synthesized noisy images, noise immunity was compared between the best-performing DNN based on the simulation study and a conventional method.

Results

The simulation study showed that a network trained by the GT of CBF and ATT in the ranges of 0 to 120 mL/100 g/min and 0 to 4500 milliseconds, respectively, had the highest performance (NMAECBF, 0.150; NRMSECBF, 0.231; CVNET CBF, 0.028; NMAEATT, 0.158; NRMSEATT, 0.257; and CVNET ATT, 0.028). Although the predicted CBF and ATT varied with the GT range of the training data sets, the appropriate settings preserved the accuracy, precision, and noise immunity of the DNN. In addition, the same results were observed in in vivo studies.

Conclusions

The GT ranges to prepare the training data affected the performance of the simulation-based supervised DNNs. The predicted CBF and ATT values depended on the GT range; inappropriate settings degraded the accuracy, whereas appropriate settings of the GT range provided accurate and precise estimates.

We aimed to investigate the effects of the ground truth (GT) ranges of CBF and ATT on the performance of the DNN when training data were prepared using arterial spin labeling signal simulation. Methods Deep neural networks were individually trained using 36 patterns of the training data sets. Simula...

03/06/2024

Current Issue: Multicenter Study of the Utility of Convolutional Neural Network and Transformer Models for the Detection and Segmentation of Meningiomas: Purpose

This study aimed to investigate the effectiveness and practicality of using models like convolutional neural network and transformer in detecting and precise segmenting meningioma from magnetic resonance images.

Methods

The retrospective study on T1-weighted and contrast-enhanced images of 523 meningioma patients from 3 centers between 2010 and 2020. A total of 373 cases split 8:2 for training and validation. Three independent test sets were built based on the remaining 150 cases. Six convolutional neural network detection models trained via transfer learning were evaluated using 4 metrics and receiver operating characteristic analysis. Detected images were used for segmentation. Three segmentation models were trained for meningioma segmentation and were evaluated via 4 metrics. In 3 test sets, intraclass consistency values were used to evaluate the consistency of detection and segmentation models with manually annotated results from 3 different levels of radiologists.

Results

The average accuracies of the detection model in the 3 test sets were 97.3%, 93.5%, and 96.0%, respectively. The model of segmentation showed mean Dice similarity coefficient values of 0.884, 0.834, and 0.892, respectively. Intraclass consistency values showed that the results of detection and segmentation models were highly consistent with those of intermediate and senior radiologists and lowly consistent with those of junior radiologists.

Conclusions

The proposed deep learning system exhibits advanced performance comparable with intermediate and senior radiologists in meningioma detection and segmentation. This system could potentially significantly improve the efficiency of the detection and segmentation of meningiomas.

images of 523 meningioma patients from 3 centers between 2010 and 2020. A total of 373 cases split 8:2 for training and validation. Three independent test sets were built based on the remaining 150 cases. Six convolutional neural network detection models trained via transfer learning were evaluated....

02/06/2024

Current Issue: Computed Tomography–Based Radiomics Signature for Predicting Segmental Chromosomal Aberrations at 1p36 and 11q23 in Pediatric Neuroblastoma: Objective

This study aimed to develop and assess the precision of a radiomics signature based on computed tomography imaging for predicting segmental chromosomal aberrations (SCAs) status at 1p36 and 11q23 in neuroblastoma.

Methods

Eighty-seven pediatric patients diagnosed with neuroblastoma and with confirmed genetic testing for SCAs status at 1p36 and 11q23 were enrolled and randomly stratified into a training set and a test set. Radiomics features were extracted from 3-phase computed tomography images and analyzed using various statistical methods. An optimal set of radiomics features was selected using a least absolute shrinkage and selection operator regression model to calculate the radiomics score for each patient. The radiomics signature was validated using receiver operating characteristic curves to obtain the area under the curve and 95% confidence interval (CI).

Results

Eight radiomics features were carefully selected and used to compute the radiomics score, which demonstrated a statistically significant distinction between the SCAs and non-SCAs groups in both sets. The radiomics signature achieved an area under the curve of 0.869 (95% CI, 0.788–0.943) and 0.883 (95% CI, 0.753–0.978) in the training and test sets, respectively. The accuracy of the radiomics signature was 0.817 and 0.778 in the training and test sets, respectively. The Hosmer-Lemeshow test confirmed that the radiomics signature was well calibrated.

Conclusions

Computed tomography–based radiomics signature has the potential to predict SCAs at 1p36 and 11q23 in neuroblastoma.

01/06/2024

Current Issue: Magnetic Resonance Imaging Features for Differentiating Low-Grade and High-Grade Malignant Peripheral Nerve Sheath Tumors: Objective

This study aimed to assess the usefulness of magnetic resonance imaging (MRI) findings for differentiating low-grade and high-grade malignant peripheral nerve sheath tumors (MPNSTs).

Methods

This study included 31 patients (onset age range, 19–83 years; mean onset age, 57 years; 9 men and 22 women) with 36 histopathologically proven MPNSTs (7 low-grade MPNSTs and 29 high-grade MPNSTs) who underwent preoperative MRI between December 2007 and October 2022. Quantitative and qualitative MRI findings were retrospectively evaluated and compared between the 2 subtypes.

Results

The maximum tumor diameter (106.1 ± 64.0 vs 54.9 ± 19.8 mm, P = 0.032) and tumor-to-muscle signal intensity ratio (SIR) of fat-suppressed gadolinium-enhanced T1-weighted images (2.69 ± 1.40 vs 1.62 ± 0.40, P = 0.005) were significantly higher in high-grade MPNSTs than in low-grade MPNSTs. The receiver operating characteristic analysis revealed that the tumor-to-muscle SIR of fat-suppressed gadolinium-enhanced T1-weighted images exhibited the highest area under the curve value (0.88), followed by the maximum tumor diameter (0.76). The sensitivity and specificity of the tumor-to-muscle SIR of fat-suppressed gadolinium-enhanced T1-weighted images for diagnosing high-grade MPNST at an optimal SIR threshold of greater than 1.73 were 90% and 83%, respectively. However, other MRI findings showed no significant differences between the 2 subtypes (P = 0.16–1.00).

Conclusions

Although the MRI findings of low-grade and high-grade MPNST overlapped considerably, the maximum tumor diameter and degree of contrast enhancement can be used to differentiate low-grade MPNST from high-grade MPNST.

9 men and 22 women) with 36 histopathologically proven MPNSTs (7 low-grade MPNSTs and 29 high-grade MPNSTs) who underwent preoperative MRI between December 2007 and October 2022. Quantitative and qualitative MRI findings were retrospectively evaluated and compared between the 2 subtypes. Results The...

31/05/2024

Current Issue: Artificial Intelligence–Based Emphysema Quantification in Routine Chest Computed Tomography: Correlation With Spirometry and Visual Emphysema Grading: Objective

The aim of the study is to assess the correlation between artificial intelligence (AI)–based low attenuation volume percentage (LAV%) with forced expiratory volume in the first second to forced vital capacity (FEV1/FVC) and visual emphysema grades in routine chest computed tomography (CT). Furthermore, optimal LAV% cutoff values for predicting a FEV1/FVC < 70% or moderate to more extensive visual emphysema grades were calculated.

Methods

In a retrospective study of 298 consecutive patients who underwent routine chest CT and spirometry examinations, LAV% was quantified using an AI-based software with a threshold < −950 HU. The FEV1/FVC was derived from spirometry, with FEV1/FVC < 70% indicating airway obstruction. The mean time interval of CT from spirometry was 3.87 ± 4.78 days. Severity of emphysema was visually graded by an experienced chest radiologist using an established 5-grade ordinal scale (Fleischner Society classification system). Spearman correlation coefficient between LAV% and FEV1/FVC was calculated. Receiver operating characteristic determined the optimal LAV% cutoff values for predicting a FEV1/FVC < 70% or a visual emphysema grade of moderate or higher (Fleischner grade 3–5).

Results

Significant correlation between LAV% and FEV1/FVC was found (ϱ = −0.477, P < 0.001). Increasing LAV% corresponded to higher visual emphysema grades. For patients with absent visual emphysema, mean LAV% was 2.98 ± 3.30, for patients with trace emphysema 3.22 ± 2.75, for patients with mild emphysema 3.90 ± 3.33, for patients with moderate emphysema 6.41 ± 3.46, for patients with confluent emphysema 9.02 ± 5.45, and for patients with destructive emphysema 16.90 ± 8.19. Optimal LAV% cutoff value for predicting a FEV1/FVC < 70 was 6.1 (area under the curve = 0.764, sensitivity = 0.773, specificity = 0.665), while for predicting a visual emphysema grade of moderate or higher, it was 4.7 (area under the curve = 0.802, sensitivity = 0.766, specificity = 0.742). Furthermore, correlation between visual emphysema grading and FEV1/FVC was found. In patients with FEV1/FVC < 70% a high proportion of subjects had emphysema grade 3 (moderate) or higher, whereas in patients with FEV1/FVC ≥ 70%, a larger proportion had emphysema grade 3 (moderate) or lower. The sensitivity for visual emphysema grading predicting a FEV1/FVC < 70% was 56.3% with an optimal cutoff point at a visual grade of 4 (confluent), demonstrating a lower sensitivity compared with LAV% (77.3%).

Conclusions

A significant correlation between AI-based LAV% and FEV1/FVC as well as visual CT emphysema grades can be found in routine chest CT suggesting that AI-based LAV% measurement might be integrated as an add-on functional parameter in the evaluation of chest CT in the future.

(CT). Furthermore, optimal LAV% cutoff values for predicting a FEV1/FVC < 70% or moderate to more extensive visual emphysema grades were calculated. Methods In a retrospective study of 298 consecutive patients who underwent routine chest CT and spirometry examinations, LAV% was quantified using an A...

30/05/2024

Current Issue: Krukenberg Tumors in Young Women: Computed Tomography and Magnetic Resonance Imaging Diagnosis: Introduction

The purpose of this report was to present the computed tomography (CT) and magnetic resonance imaging (MRI) features of Krukenberg tumors and to review the pertinent clinical data about the rising incidence of this malignancy among young women.

Material and Methods

This series included 8 women who ranged in age from 24 to 44 years (mean, 36.3 years). They were diagnosed to have Krukenberg tumors during a 5-year period (2011–2016). All patients were evaluated by abdominal CT and pelvic or transvaginal sonography. Five of them also had MRI of the abdomen, and 3 had positron emission tomography scans.

Results

The primary cancer was located in the stomach of 7 patients and in the colon in 1. The initial presentation was due to large pelvic mass and abdominal distention by ascites in 3 patients, gastrointestinal symptoms in 4, and small bowel obstruction by carcinoma of the ascending colon in 1 woman. Ovarian metastases were demonstrated on the initial imaging examination of 5 patients and developed as metachronous lesion during follow-up in 3 cases. Six women died 3 to 23 months (mean, 11 months) after the diagnosis. One has survived for 6 years after extensive surgery, and 1 was lost to follow-up.

Conclusions

Krukenberg tumors are being diagnosed with an increasing frequency because of well-documented rising incidence of gastric and colore**al carcinomas among young women. Therefore, those presenting with gastrointestinal cancer should have careful imaging of their ovaries for possible metastases, and conversely, the clinical or sonographic detection of bilateral ovarian masses would require evaluation by CT or MRI of the abdomen in search for a potential primary gastrointestinal cancer. This report highlights the CT and MRI features of this neoplastic process and reviews the current concepts about its development and management.

s included 8 women who ranged in age from 24 to 44 years (mean, 36.3 years). They were diagnosed to have Krukenberg tumors during a 5-year period (2011–2016). All patients were evaluated by abdominal CT and pelvic or transvaginal sonography. Five of them also had MRI of the abdomen, and 3 had posi...

29/05/2024

Current Issue: The Clinical Value of Apparent Diffusion Coefficient of Readout Segmentation of Long Variable Echo Trains and Correlation With Ki-67 Expression in Distal Re**al Cancer: Objective

The aim of the study is to explore the clinical value of the apparent diffusion coefficient (ADC) derived from the readout segmentation of long variable echo trains (RESOLVE) technique for identifying clinicopathologic features of distal re**al cancer and correlations between ADC and Ki-67 expression.

Methods

The data of 112 patients with a proven pathology of distal re**al cancer who underwent preoperative magnetic resonance imaging were retrospectively analyzed. The mean ADC value was measured using the “full-layer and center” method. Differences in ADC values and Ki-67 expression in different clinical stages, pathological types, and tumor differentiation were compared using analysis of variance. Correlations between ADC value and clinicopathologic features were assessed using Spearman correlation analysis.

Results

Interobserver agreement of confidence levels from 2 radiologists was excellent for ADC measurement (k = 0.85). Patients with a lower clinical stage, well-differentiated adenocarcinomas, and a higher possibility of mucinous adenocarcinoma exhibited a positive correlation with higher ADC values, but these factors were negatively correlated with Ki-67 expression (all P < 0.05). We found that ADC value was negatively correlated with Ki-67 expression (r = −0.62, P < 0.001).

Conclusions

The ADC value generated by RESOLVE sequences was significantly associated with clinicopathologic features and Ki-67 expression in patients with distal re**al cancer in this study. Thus, the ADC value could be considered a new noninvasive imaging biomarker that could be helpful in predicting the biological properties of distal re**al cancer.

Ki-67 expression. Methods The data of 112 patients with a proven pathology of distal re**al cancer who underwent preoperative magnetic resonance imaging were retrospectively analyzed. The mean ADC value was measured using the “full-layer and center” method. Differences in ADC values and Ki-67 ex...

28/05/2024

Current Issue: Use of Computed Tomography–Based Texture Analysis to Differentiate Benign From Malignant Salivary Gland Lesions: Objective

Salivary gland lesions show overlapping morphological findings and types of time/intensity curves. This research aimed to evaluate the role of 2-phase multislice spiral computed tomography (MSCT) texture analysis in differentiating between benign and malignant salivary gland lesions.

Methods

In this prospective study, MSCT was carried out on 90 patients. Each lesion was segmented on axial computed tomography (CT) images manually, and 33 texture features and morphological CT features were assessed. Logistic regression analysis was used to confirm predictors of malignancy (P < 0.05 was considered to be statistically significant), followed by receiver operating characteristics analysis to assess the diagnostic performance.

Results

Univariate logistic regression analysis revealed that morphological CT features (shape, size, and invasion of adjacent tissues) and 17 CT texture parameters had significant differences between benign and malignant lesions (P < 0.05). Multivariate binary logistic regression demonstrated that shape, invasion of adjacent tissues, entropy, and inverse difference moment were independent factors for malignant tumors. The diagnostic accuracy values of multivariate binary logistic models based on morphological parameters, CT texture features, and a combination of both were 87.8%, 90%, and 93.3%, respectively.

Conclusions

Two-phase MSCT texture analysis was conducive to differentiating between malignant and benign neoplasms in the salivary gland, especially when combined with morphological CT features.

Methods In this prospective study, MSCT was carried out on 90 patients. Each lesion was segmented on axial computed tomography (CT) images manually, and 33 texture features and morphological CT features were assessed. Logistic regression analysis was used to confirm predictors of malignancy (P < 0.0...

27/05/2024

Current Issue: Patients With Post–COVID-19 Respiratory Condition: Chest Computed Tomography Findings and Pulmonary Function Tests and Comparison With Asymptomatic Participants: Objective

The aims of this study were to assess the chest computed tomography (CT) findings in post–COVID-19 respiratory condition (rPCC) patients and compare the findings with asymptomatic participants (APs). It also aimed to evaluate the relationship between CT findings and pulmonary function tests (PFTs) in rPCC patients. Finally, it aimed to compare the quantitative chest CT findings and PFT results of patients with rPCC and APs.

Methods

We retrospectively enrolled consecutive patients with rPCC who underwent unenhanced chest CT and PFTs between June 2020 and September 2022. In addition, a control group (APs) was prospectively formed and underwent nonenhanced chest CT and PFTs. The presence and extent of abnormalities in unenhanced chest CT images were evaluated qualitatively and semiquantitatively in a blinded manner. We used fully automatic software for automatic lung and airway segmentation and quantitative analyses.

Results

Sixty-three patients with rPCC and 23 APs were investigated. Reticulation/interstitial thickening and extent of parenchymal abnormalities on CT were significantly greater in the rPCC group than in the control group (P = 0.001 and P = 0.004, respectively). Computed tomography extent score was significantly related to length of hospital stay, age, and intensive care unit stay (all Ps ≤ 0.006). The rPCC group also had a lower 85th percentile attenuation lung volume (P = 0.037). The extent of parenchymal abnormalities was significantly correlated with carbon monoxide diffusing capacity (r = −0.406, P = 0.001), forced vital capacity (FVC) (r = −0.342, P = 0.002), and forced expiratory volume in 1 second/FVC (r = 0.427, P < 0.001) values. Pulmonary function tests revealed significantly lower carbon monoxide diffusing capacity (P < 0.001), FVC (P = 0.036), and total lung capacity (P < 0.001) values in the rPCC group.

Conclusions

The rPCC is characterized by impaired PFTs, a greater extent of lung abnormalities on CT, and decreased 85th percentile attenuation lung volume. Advanced age, intensive care unit admission history, and extended hospital stay are risk factors for chest CT abnormalities.

n tests (PFTs) in rPCC patients. Finally, it aimed to compare the quantitative chest CT findings and PFT results of patients with rPCC and APs. Methods We retrospectively enrolled consecutive patients with rPCC who underwent unenhanced chest CT and PFTs between June 2020 and September 2022. In addit...

26/05/2024

Current Issue: Characterization of Demographical Histologic Diversity in Small Renal Masses With the Clear Cell Likelihood Score: Objective

This study aimed to develop a diagnostic model to estimate the distribution of small renal mass (SRM; ≤4 cm) histologic subtypes for patients with different demographic backgrounds and clear cell likelihood score (ccLS) designations.

Materials and Methods

A bi-institution retrospective cohort study was conducted where 347 patients (366 SRMs) underwent magnetic resonance imaging and received a ccLS before pathologic confirmation between June 2016 and November 2021. Age, s*x, race, ethnicity, socioeconomic status, body mass index (BMI), and the ccLS were tabulated. The socioeconomic status for each patient was determined using the Area Deprivation Index associated with their residential address. The magnetic resonance imaging–derived ccLS assists in the characterization of SRMs by providing a likelihood of clear cell renal cell carcinoma (ccRCC). Pathological subtypes were grouped into four categories (ccRCC, papillary renal cell carcinoma, other renal cell carcinomas, or benign). Generalized estimating equations were used to estimate probabilities of the pathological subtypes across different patient subgroups.

Results

Race and ethnicity, BMI, and ccLS were significant predictors of histology (all P < 0.001). Obese (BMI, ≥30 kg/m2) Hispanic patients with ccLS of ≥4 had the highest estimated rate of ccRCC (97.1%), and normal-weight (BMI,

pective cohort study was conducted where 347 patients (366 SRMs) underwent magnetic resonance imaging and received a ccLS before pathologic confirmation between June 2016 and November 2021. Age, s*x, race, ethnicity, socioeconomic status, body mass index (BMI), and the ccLS were tabulated. The socio...

25/05/2024

Current Issue: Three-dimensional Volumetric Visceral and Subcutaneous Fat Analysis on Opportunistic Computed Tomography Imaging of Patients With Greater Trochanteric Pain Syndrome Compared With Those With Predominant Osteoarthritis: A Case-Control Study: Objective

This study aimed to address the gap in knowledge assessing the impact of visceral and subcutaneous body fat on 3-dimensional computed tomography imaging in patients with greater trochanteric pain syndrome (GTPS) in comparison with those primarily diagnosed with osteoarthritis (OA).

Materials and Methods

We evaluated adult patients with a confirmed diagnosis of GTPS from our institutional hip-preservation clinic spanning 2011 to 2022. Selection criteria included their initial clinic visit for hip pain and a concurrent pelvis computed tomography scan. These patients were age- and s*x-matched to mild-moderate OA patients selected randomly from the database. Visceral and subcutaneous fat areas were measured volumetrically from the sacroiliac joint to the lesser trochanter using an independent software. Interreader reliability was also calculated.

Results

A total of 93 patients met the study criteria, of which 37 belonged to the GTPS group and 56 belonged to the OA group. Both groups were s*x and race matched. Average age in GTPS and OA groups was 59.3 years and 56 years, respectively. For GTPS group, average body mass index was 28.9 kg/m2, and for the OA group, average body mass index was 29.9 kg/m2, with no significant difference (P > 0.05). Two-sample t test showed no significant differences in the visceral fat, subcutaneous fat, or the visceral fat to total fat volume ratio between the GTPS and OA groups. There was excellent interreader reliability.

Conclusions

Our results indicate that there is no significant difference in fat distribution and volumes among GTPS and OA patients. This suggests that being overweight or obese may not be directly linked or contribute to the onset of GTPS. Other factors, such as gluteal tendinopathy, bursitis, or iliotibial band syndrome, might be responsible and need further investigation.

Materials and Methods We evaluated adult patients with a confirmed diagnosis of GTPS from our institutional hip-preservation clinic spanning 2011 to 2022. Selection criteria included their initial clinic visit for hip pain and a concurrent pelvis computed tomography scan. These patients were age- an...

24/05/2024

Current Issue: Prototype Description and Ex Vivo Evaluation of a System for Combined Endore**al Magnetic Resonance Imaging and In-Bore Biopsy of the Prostate: We describe early ex vivo proof-of-concept testing of a novel system composed of a disposable endore**al coil and converging multichannel needle guide with a reusable clamp stand, embedded electronics, and baseplate to allow for endore**al magnetic resonance (MR) imaging and in-bore MRI-targeted biopsy of the prostate as a single integrated procedure. Using prostate phantoms imaged with standard T2-weighted sequences in a Siemens 3T Prisma MR scanner, we measured the signal-to-noise ratio in successive 1-cm distances from the novel coil and from a commercially available inflatable balloon coil and measured the lateral and longitudinal deviation of the tip of a deployed MR compatible needle from the intended target point. Signal-to-noise ratio obtained with the novel system was significantly better than the inflatable balloon coil at each of five 1-cm intervals, with a mean improvement of 78% (P < 0.05). In a representative sampling of 15 guidance channels, the mean lateral deviation for MR imaging–guided needle positioning was 1.7 mm and the mean longitudinal deviation was 2.0 mm. Our ex vivo results suggest that our novel system provides significantly improved signal-to-noise ratio when compared with an inflatable balloon coil and is capable of accurate MRI-guided needle deployment.

biopsy of the prostate as a single integrated procedure. Using prostate phantoms imaged with standard T2-weighted sequences in a Siemens 3T Prisma MR scanner, we measured the signal-to-noise ratio in successive 1-cm distances from the novel coil and from a commercially available inflatable balloon c...

23/05/2024

Current Issue: Noninvasive Isocitrate Dehydrogenase 1 Status Prediction in Grade II/III Glioma Based on Magnetic Resonance Images: A Transfer Learning Strategy: Objective

The aim of this study was to evaluate transfer learning combined with various convolutional neural networks (TL-CNNs) in predicting isocitrate dehydrogenase 1 (IDH1) status of grade II/III gliomas.

Methods

Grade II/III glioma patients diagnosed at the Tangdu Hospital (August 2009 to May 2017) were retrospectively enrolled, including 54 patients with IDH1 mutant and 56 patients with wild-type IDH1. Convolutional neural networks, AlexNet, GoogLeNet, ResNet, and VGGNet were fine-tuned with T2-weighted imaging (T2WI), fluid attenuation inversion recovery (FLAIR), and contrast-enhanced T1-weighted imaging (T1CE) images. The single-modal networks were integrated with averaged sigmoid probabilities, logistic regression, and support vector machine. FLAIR-T1CE-fusion (FC-fusion), T2WI-T1CE-fusion (TC-fusion), and FLAIR-T2WI-T1CE-fusion (FTC-fusion) were used for fine-tuning TL-CNNs.

Results

IDH1-mutant prediction accuracies using AlexNet, GoogLeNet, ResNet, and VGGNet achieved 70.0% (AUC = 0.660), 65.0% (AUC = 0.600), 70.0% (AUC = 0.700), and 80.0% (AUC = 0.730) for T2WI images, 70.0% (AUC = 0.660), 70.0% (AUC = 0.620), 70.0% (AUC = 0.710), and 80.0% (AUC = 0.720) for FLAIR images, and 73.7% (AUC = 0.744), 73.7% (AUC = 0.656), 73.7% (AUC = 0.633), and 73.7% (AUC = 0.700) for T1CE images, respectively. The highest AUC (0.800) was achieved using VGGNet and FC-fusion images.

Conclusions

TL-CNNs (especially VGGNet) had a potential predictive value for IDH1-mutant status of grade II/III gliomas.

9 to May 2017) were retrospectively enrolled, including 54 patients with IDH1 mutant and 56 patients with wild-type IDH1. Convolutional neural networks, AlexNet, GoogLeNet, ResNet, and VGGNet were fine-tuned with T2-weighted imaging (T2WI), fluid attenuation inversion recovery (FLAIR), and contrast-...