Preoperative Differentiation of HER2-Zero and HER2-Low from HER2-Positive Invasive Ductal Breast Cancers Using BI-RADS MRI Features and Machine Learning Modeling.

BACKGROUND: Accurate determination of human epidermal growth factor receptor 2 (HER2) is important for choosing optimal HER2 targeting treatment strategies. HER2-low is currently considered HER2-negative, but patients may be eligible to receive new anti-HER2 drug conjugates. PURPOSE: To use breast MRI BI-RADS features for classifying three HER2 levels, first to distinguish HER2-zero from HER2-low/positive (Task-1), and then to distinguish HER2-low from HER2-positive (Task-2). STUDY TYPE: Retrospective. POPULATION: 621 invasive ductal cancer, 245 HER2-zero, 191 HER2-low, and 185 HER2-positive. For Task-1, 488 cases for training and 133 for testing. For Task-2, 294 cases for training and 82 for testing. FIELD STRENGTH/SEQUENCE: 3.0 T; 3D T1-weighted DCE, short time inversion recovery T2, and single-shot EPI DWI. ASSESSMENT: Pathological information and BI-RADS features were compared. Random Forest was used to select MRI features, and then four machine learning (ML) algorithms: decision tree (DT), support vector machine (SVM), k-nearest neighbors (k-NN), and artificial neural nets (ANN), were applied to build models. STATISTICAL TESTS: Chi-square test, one-way analysis of variance, and Kruskal-Wallis test were performed. The P values <0.05 were considered statistically significant. For ML models, the generated probability was used to construct the ROC curves. RESULTS: Peritumoral edema, the presence of multiple lesions and non-mass enhancement (NME) showed significant differences. For distinguishing HER2-zero from non-zero (low + positive), multiple lesions, edema, margin, and tumor size were selected, and the k-NN model achieved the highest AUC of 0.86 in the training set and 0.79 in the testing set. For differentiating HER2-low from HER2-positive, multiple lesions, edema, and margin were selected, and the DT model achieved the highest AUC of 0.79 in the training set and 0.69 in the testing set. DATA CONCLUSION: BI-RADS features read by radiologists from preoperative MRI can be analyzed using more sophisticated feature selection and ML algorithms to build models for the classification of HER2 status and identify HER2-low. TECHNICAL EFFICACY: Stage 2.

Radiomics analysis for the prediction of locoregional recurrence of locally advanced oropharyngeal cancer and hypopharyngeal cancer.

PURPOSE: By radiomic analysis of the postcontrast CT images, this study aimed to predict locoregional recurrence (LR) of locally advanced oropharyngeal cancer (OPC) and hypopharyngeal cancer (HPC). METHODS: A total of 192 patients with stage III-IV OPC or HPC from two independent cohort were randomly split into a training cohort with 153 cases and a testing cohort with 39 cases. Only primary tumor mass was manually segmented. Radiomic features were extracted using PyRadiomics, and then the support vector machine was used to build the radiomic model with fivefold cross-validation process in the training data set. For each case, a radiomics score was generated to indicate the probability of LR. RESULTS: There were 94 patients with LR assigned in the progression group and 98 patients without LR assigned in the stable group. There was no significant difference of TNM staging, treatment strategies and common risk factors between these two groups. For the training data set, the radiomics model to predict LR showed 83.7% accuracy and 0.832 (95% CI 0.72, 0.87) area under the ROC curve (AUC). For the test data set, the accuracy and AUC slightly declined to 79.5% and 0.770 (95% CI 0.64, 0.80), respectively. The sensitivity/specificity of training and test data set for LR prediction were 77.6%/89.6%, and 66.7%/90.5%, respectively. CONCLUSIONS: The image-based radiomic approach could provide a reliable LR prediction model in locally advanced OPC and HPC. Early identification of those prone to post-treatment recurrence would be helpful for appropriate adjustments to treatment strategies and post-treatment surveillance.

Magnetic Resonance Imaging Features Associated with a High and Low Expression of Tumor-Infiltrating Lymphocytes: A Stratified Analysis According to Molecular Subtypes.

A total of 457 patients, including 241 HR+/HER2- patients, 134 HER2+ patients, and 82 TN patients, were studied. The percentage of TILs in the stroma adjacent to the tumor cells was assessed using a 10% cutoff. The low TIL percentages were 82% in the HR+ patients, 63% in the HER2+ patients, and 56% in the TN patients (p < 0.001). MRI features such as morphology as mass or non-mass enhancement (NME), shape, margin, internal enhancement, presence of peritumoral edema, and the DCE kinetic pattern were assessed. Tumor sizes were smaller in the HR+/HER2- group (p < 0.001); HER2+ was more likely to present as NME (p = 0.031); homogeneous enhancement was mostly seen in HR+ (p < 0.001); and the peritumoral edema was present in 45% HR+, 71% HER2+, and 80% TN (p < 0.001). In each subtype, the MR features between the high- vs. low-TIL groups were compared. In HR+/HER2-, peritumoral edema was more likely to be present in those with high TILs (70%) than in those with low TILs (40%, p < 0.001). In TN, those with high TILs were more likely to present a regular shape (33%) than those with low TILs (13%, p = 0.029) and more likely to present the circumscribed margin (19%) than those with low TILs (2%, p = 0.009).

Pretreatment diffusion-weighted imaging for prediction of relapsed and refractory primary central nervous system lymphoma.

Objectives: A subset of primary central nervous system lymphoma (PCNSL) has been shown to undergo an early relapsed/refractory (R/R) period after first-line chemotherapy. This study investigated the pretreatment clinical and MRI features to predict R/R in PCNSL, emphasizing the apparent diffusion coefficient (ADC) values in diffusion-weighted imaging (DWI). Methods: This retrospective study investigated the pretreatment MRI features for predicting R/R in PCNSL. Only patients who had undergone complete preoperative and postoperative MRI follow-up studies were included. From January 2006 to December 2021, 52 patients from two medical institutions with a diagnosis of PCNSL were included (median follow-up time, 26.3 months). Among these, 24 (46.2%) had developed R/R (median time to relapse, 13 months). Cox proportional hazard regression analyses were performed to determine hazard ratios for all parameters. Results: Significant predictors of R/R in PCNSL were female sex, complete response (CR) to first-line chemotherapy, and ADC value/ratio (p < 0.05). Cut-off points of ADC values and ADC ratios for prediction of R/R were 0.68 × 10-3 mm2/s and 0.97, with AUCs of 0.78 and 0.77, respectively (p < 0.05). Multivariate Cox proportional hazards analysis showed that failure of CR to first-line chemotherapy and low ADC values (<0.68 × 10-3 mm2/s) were significant risk factors for R/R, with hazard ratios of 5.22 and 14.45, respectively (p < 0.05). Kaplan-Meier analysis showed that lower ADC values and ratios predicted significantly shorter progression-free survival (p < 0.05). Conclusion: Pretreatment ADC values in DWI offer quantitative valuable information for the treatment planning in PCNSL.

Deep Learning-based Automatic Diagnosis of Breast Cancer on MRI Using Mask R-CNN for Detection Followed by ResNet50 for Classification.

RATIONALE AND OBJECTIVES: Diagnosis of breast cancer on MRI requires, first, the identification of suspicious lesions; second, the characterization to give a diagnostic impression. We implemented Mask Reginal-Convolutional Neural Network (R-CNN) to detect abnormal lesions, followed by ResNet50 to estimate the malignancy probability. MATERIALS AND METHODS: Two datasets were used. The first set had 176 cases, 103 cancer, and 73 benign. The second set had 84 cases, 53 cancer, and 31 benign. For detection, the pre-contrast image and the subtraction images of left and right breasts were used as inputs, so the symmetry could be considered. The detected suspicious area was characterized by ResNet50, using three DCE parametric maps as inputs. The results obtained using slice-based analyses were combined to give a lesion-based diagnosis. RESULTS: In the first dataset, 101 of 103 cancers were detected by Mask R-CNN as suspicious, and 99 of 101 were correctly classified by ResNet50 as cancer, with a sensitivity of 99/103 = 96%. 48 of 73 benign lesions and 131 normal areas were identified as suspicious. Following classification by ResNet50, only 16 benign and 16 normal areas remained as malignant. The second dataset was used for independent testing. The sensitivity was 43/53 = 81%. Of the total of 121 identified non-cancerous lesions, only 6 of 31 benign lesions and 22 normal tissues were classified as malignant. CONCLUSION: ResNet50 could eliminate approximately 80% of false positives detected by Mask R-CNN. Combining Mask R-CNN and ResNet50 has the potential to develop a fully-automatic computer-aided diagnostic system for breast cancer on MRI.

Prediction of poor outcome in stroke patients using radiomics analysis of intraparenchymal and intraventricular hemorrhage and clinical factors.

PURPOSE: To build three prognostic models using radiomics analysis of the hemorrhagic lesions, clinical variables, and their combination, to predict the outcome of stroke patients with spontaneous intracerebral hemorrhage (sICH). MATERIALS AND METHODS: Eighty-three sICH patients were included. Among them, 40 patients (48.2%) had poor prognosis with modified Rankin scale (mRS) of 5 and 6 at discharge, and the prognostic model was built to differentiate mRS ≤ 4 vs. 5 + 6. The region of interest (ROI) of intraparenchymal hemorrhage (IPH) and intraventricular hemorrhage (IVH) were separately segmented. Features were extracted using PyRadiomics, and the support vector machine was applied to select features and build radiomics models based on IPH and IPH + IVH. The clinical models were built using multivariate logistic regression, and then the radiomics scores were combined with clinical variables to build the combined model. RESULTS: When using IPH, the AUC for radiomics, clinical, and combined model was 0.78, 0.82, and 0.87, respectively. When using IPH + IVH, the AUC was increased to 0.80, 0.84, and 0.90, respectively. The combined model had a significantly improved AUC compared to the radiomics by DeLong test. A clinical prognostic model based on the ICH score of 0-1 only achieved AUC of 0.71. CONCLUSIONS: The combined model using the radiomics score derived from IPH + IVH and the clinical factors could achieve a high accuracy in prediction of sICH patients with poor outcome, which may be used to assist in making the decision about the optimal care.

Non-Contrast-Enhanced MR Arteriography of Potential Living-Related Liver Donor: Using Contrast Enhanced CT Arteriography as Standard Reference.

BACKGROUND: Contrast-enhanced computed tomography angiography (CTA) and magnetic resonance angiography (MRA) are the primary modalities to assess donors' vessels before transplant surgery. Radiation and contrast medium are potentially harmful to donors. PURPOSE: To compare the image quality and visualization scores of hepatic arteries on CTA and balanced steady-state free-precession (bSSFP) non-contrast-enhanced MRA (NC-MRA), and to evaluate if bSSFP NC-MRA can potentially be a substitute for CTA. STUDY TYPE: Prospective. POPULATION: Fifty-six consecutive potential living-related liver donors (30.9 ± 8.4 years; 31 men). FIELD STRENGTH/SEQUENCE: 1.5T; four bSSFP NC-MRA sequences: respiratory-triggered (Inhance inflow inversion recovery [IFIR]) and three breath-hold (BH); and CTA. ASSESSMENT: The artery-to-liver contrast (Ca-l) was quantified. Three radiologists independently assigned visualization scores using a four-point scale to potential origins, segments, and branches of the hepatic arteries, determined the anatomical variants based on Hiatt's classification, and assessed the image quality of NC-MRA sequences. STATISTICAL TESTS: Fleiss' kappa to evaluate the readers' agreement. Repeat measured ANOVA or Friedman test to compare Ca-l of each NC-MRA. Friedman test to compare overall image quality and visualization scores; post hoc analysis using Wilcoxon signed-rank test. P-value <0.05 was considered statistically significant. RESULTS: Inhance IFIR Ca-l was significantly higher than all BH bSSFP Ca-l (0.56 [0.45-0.64] vs. 0.37 [0.29-0.47] to 0.41 [0.23-0.51]). Overall image quality score of BH bSSFP TI1200 was significantly higher than other NC-MRA (4 [4-4] vs. 4 [3 to 4-4]). The median visualization scores of almost all arteries on CTA were significantly higher than on NC-MRA (4 [3 to 4-4] vs. 1 [1-2] to 4 [4-4]). The median visualization scores were all 4 [4-4 ] on Inhance IFIR with >92.3% observed scores ≥3, except the segment 4 branch (3 [1-4], 53.6%). The identification rates of arterial variants were 92.9%-97% on Inhance IFIR. DATA CONCLUSIONS: Although CTA is superior to the NC-MRA, all NC-MRA depict the donor arterial anatomy well. Inhance IFIR can potentially be an alternative image modality for CTA to evaluate the arterial variants of living donors. LEVEL OF EVIDENCE: 3 TECHNICAL EFFICACY: Stage 2.

The Added Value of Intraventricular Hemorrhage on the Radiomics Analysis for the Prediction of Hematoma Expansion of Spontaneous Intracerebral Hemorrhage.

Background: Among patients undergoing head computed tomography (CT) scans within 3 h of spontaneous intracerebral hemorrhage (sICH), 28% to 38% have hematoma expansion (HE) on follow-up CT. This study aimed to predict HE using radiomics analysis and investigate the impact of intraventricular hemorrhage (IVH) compared with the conventional approach based on intraparenchymal hemorrhage (IPH) alone. Methods: This retrospective study enrolled 127 patients with baseline and follow-up non-contrast CT (NCCT) within 4~72 h of sICH. IPH and IVH were outlined separately for performing radiomics analysis. HE was defined as an absolute hematoma growth > 6 mL or percentage growth > 33% of either IPH (HEP) or a combination of IPH and IVH (HEP+V) at follow-up. Radiomic features were extracted using PyRadiomics, and then the support vector machine (SVM) was used to build the classification model. For each case, a radiomics score was generated to indicate the probability of HE. Results: There were 57 (44.9%) HEP and 70 (55.1%) non-HEP based on IPH alone, and 58 (45.7%) HEP+V and 69 (54.3%) non-HEP+V based on IPH + IVH. The majority (>94%) of HE patients had poor early outcomes (death or modified Rankin Scale > 3 at discharge). The radiomics model built using baseline IPH to predict HEP (RMP) showed 76.4% accuracy and 0.73 area under the ROC curve (AUC). The other model using IPH + IVH to predict HEP+V (RMP+V) had higher accuracy (81.9%) with AUC = 0.80, and this model could predict poor outcomes. The sensitivity/specificity of RMP and RMP+V for HE prediction were 71.9%/80.0% and 79.3%/84.1%, respectively. Conclusion: The proposed radiomics approach with additional IVH information can improve the accuracy in prediction of HE, which is associated with poor clinical outcomes. A reliable radiomics model may provide a robust tool to help manage ICH patients and to enroll high-risk ICH cases into anti-expansion or neuroprotection drug trials.

Intravoxel Incoherent Motion-Diffusion-Weighted MRI for Investigation of Delayed Graft Function Immediately after Kidney Transplantation.

Purpose: A non-invasive way of assessing post-transplant renal graft function has been needed. This study aimed to assess the micro-structural and micro-functional status of graft kidneys by using intravoxel incoherent motion- (IVIM-) diffusion-weighted imaging (DWI) to investigate delayed graft function (DGF) immediately after transplantation. Method: A prospective study was conducted on 37 patients, 14 with early graft function (EGF) and 23 with DGF (9 with complication, 14 without) who underwent IVIM-DWI, most often within 1-7 days after kidney transplantation. A total of 37 cases were collected and all the participants have been well-informed and signed their consents. In addition, the study conducted in this paper was approved by the Ethics Committee of Clinical Research, Taichung Veterans General Hospital (IRB number: CE14065). Using biexponential analysis of slow diffusion coefficient (D slow), fast diffusion coefficient (D fast), and perfusion fraction was performed. The apparent diffusion coefficient (ADC) was calculated by use of a monoexponential model. All parameters were measured from three different regions-of-interest (ROI), covering the entire renal parenchyma, cortex, and medulla. Results: D slow, perfusion fraction, and ADC were significantly higher in patients with EGF than DGF (all p values values <0.001). Especially, ADC measured from ROI covering the entire kidney parenchyma had the best cut-off value (1.93µm2/msec) with the highest area under the receiver operating characteristic curve (AUC 0.943) in differentiating EGF from DGF. For analysis of pair-wise differences, only the perfusion fraction values, measured from the ROI covering the renal cortex, were significantly higher in 14 DGF patients with no complications than in the 9 DGF patients with complications, with the best cut-off value of 12.3% and the AUC of 0.844. Conclusion: Noninvasive IVIM-DWI reliably differentiates DGF from EGF after kidney transplantation, and it may aid in identifying posttransplant complications and indications for renal biopsy.

Prediction of Intraparenchymal Hemorrhage Progression and Neurologic Outcome in Traumatic Brain Injury Patients Using Radiomics Score and Clinical Parameters.

(1) Background: Radiomics analysis of spontaneous intracerebral hemorrhages on computed tomography (CT) images has been proven effective in predicting hematoma expansion and poor neurologic outcome. In contrast, there is limited evidence on its predictive abilities for traumatic intraparenchymal hemorrhage (IPH). (2) Methods: A retrospective analysis of 107 traumatic IPH patients was conducted. Among them, 45 patients (42.1%) showed hemorrhagic progression of contusion (HPC) and 51 patients (47.7%) had poor neurological outcome. The IPH on the initial CT was manually segmented for radiomics analysis. After feature extraction, selection and repeatability evaluation, several machine learning algorithms were used to derive radiomics scores (R-scores) for the prediction of HPC and poor neurologic outcome. (3) Results: The AUCs for R-scores alone to predict HPC and poor neurologic outcome were 0.76 and 0.81, respectively. Clinical parameters were used to build comparison models. For HPC prediction, variables including age, multiple IPH, subdural hemorrhage, Injury Severity Score (ISS), international normalized ratio (INR) and IPH volume taken together yielded an AUC of 0.74, which was significantly (p = 0.022) increased to 0.83 after incorporation of the R-score in a combined model. For poor neurologic outcome prediction, clinical variables of age, Glasgow Coma Scale, ISS, INR and IPH volume showed high predictability with an AUC of 0.92, and further incorporation of the R-score did not improve the AUC. (4) Conclusion: The results suggest that radiomics analysis of IPH lesions on initial CT images has the potential to predict HPC and poor neurologic outcome in traumatic IPH patients. The clinical and R-score combined model further improves the performance of HPC prediction.

Deep Learning for Prediction of Progression and Recurrence in Nonfunctioning Pituitary Macroadenomas: Combination of Clinical and MRI Features.

Objectives: A subset of non-functioning pituitary macroadenomas (NFMAs) may exhibit early progression/recurrence (P/R) after tumor resection. The purpose of this study was to apply deep learning (DL) algorithms for prediction of P/R in NFMAs. Methods: From June 2009 to December 2019, 78 patients diagnosed with pathologically confirmed NFMAs, and who had undergone complete preoperative MRI and postoperative MRI follow-up for more than one year, were included. DL classifiers including multi-layer perceptron (MLP) and convolutional neural network (CNN) were used to build predictive models. Categorical and continuous clinical data were fed into the MLP model, and images of preoperative MRI (T2WI and contrast enhanced T1WI) were analyzed by the CNN model. MLP, CNN and multimodal CNN-MLP architectures were performed to predict P/R in NFMAs. Results: Forty-two (42/78, 53.8%) patients exhibited P/R after surgery. The median follow-up time was 42 months, and the median time to P/R was 25 months. As compared with CNN using MRI (accuracy 83%, precision 87%, and AUC 0.84) or MLP using clinical data (accuracy 73%, precision 73%, and AUC 0.73) alone, the multimodal CNN-MLP model using both clinical and MRI features showed the best performance for prediction of P/R in NFMAs, with accuracy 83%, precision 90%, and AUC 0.85. Conclusions: DL architecture incorporating clinical and MRI features performs well to predict P/R in NFMAs. Pending more studies to support the findings, the results of this study may provide valuable information for NFMAs treatment planning.

Preoperative Apparent Diffusion Coefficient Values for Differentiation between Low and High Grade Meningiomas: An Updated Systematic Review and Meta-Analysis.

The meta-analysis aimed to compare the preoperative apparent diffusion coefficient (ADC) values between low-grade meningiomas (LGMs) and high-grade meningiomas (HGMs). Medline, Cochrane, Scopus, and Embase databases were screened up to January 2022 for studies investigating the ADC values of meningiomas. The study endpoint was the reported ADC values for LGMs and HGMs. Further subgroup analyses between 1.5T and 3T MRI scanners, ADC threshold values, ADC in different histological LGMs, and correlation coefficients (r) between ADC and Ki-67 were also performed. The quality of studies was evaluated by the quality assessment of diagnostic accuracy studies (QUADAS-2). A χ2-based test of homogeneity was performed using Cochran's Q statistic and inconsistency index (I2). Twenty-five studies with a total of 1552 meningiomas (1102 LGMs and 450 HGMs) were included. The mean ADC values (×10−3 mm2/s) were 0.92 and 0.79 for LGMs and HGMs, respectively. Compared with LGMs, significantly lower mean ADC values for HGMs were observed with a pooled difference of 0.13 (p < 0.00001). The results were consistent in both 1.5T and 3T MRI scanners. For ADC threshold values, pooled sensitivity of 69%, specificity of 82%, and AUC of 0.84 are obtained for differentiation between LGMs and HGMs. The mean ADC (×10−3 mm2/s) in different histological LGMs ranged from 0.87 to 1.22. Correlation coefficients (r) of mean ADC and Ki-67 ranged from −0.29 to −0.61. Preoperative ADC values are a useful tool for differentiating between LGMs and HGMs. Results of this study provide valuable information for planning treatments in meningiomas.

Usage of image registration and three-dimensional visualization tools on serial computed tomography for the analysis of patients with traumatic intraparenchymal hemorrhages.

The aim of this study was to apply registration and three-dimensional (3D) display tools to assess the evolution of intraparenchymal hemorrhage (IPH) in patients with traumatic brain injury (TBI). We identified 109 TBI patients who had two computed tomography (CT) scans within 4 days retrospectively. The IPH was manually outlined. The registration was performed in 39 lesions from 29 patients with lesion volume < 1.5 cm on both baseline and follow-up CT. The center of mass (COM) of each lesion was calculated, and the distance between baseline and follow-up CT was used to evaluate the registration effect. The mean distances of COM before registration in the XYZ, XY, and YZ coordinates were 20.5 ± 10.2 mm, 17.8 ± 9.4 mm, and 15.9 ± 9.4 mm, respectively, which decreased significantly (p < 0.001) to 7.9 ± 4.9, 7.8 ± 5.0, and 6.1 ± 4.1 mm after registration. A 3D short video displaying the rendering view of all lesions in 34 randomly selected patients from baseline and follow-up scans were presented side-by-side for comparison. The detection rate of new IPH lesions increased in 3D videos (100%) as compared with axial CT slices (78.6-92.9%). A very high interrater agreement (k = 0.856) on perceiving IPH lesion progression upon viewing 3D video was noted, and the absolute volume increase was significantly higher (p < 0.001) for progressive lesions (median 7.36 cc) over non-progressive lesions (median 0.01 cc). Compared to patients with spontaneous hemorrhagic stroke, evaluation of multiple small traumatic hemorrhages in TBI is more challenging. The applied image analysis and visualization methods may provide helpful tools for comparing changes between serial CT scans.

A deep learning-based method for the diagnosis of vertebral fractures on spine MRI: retrospective training and validation of ResNet.

PURPOSE: To improve the performance of less experienced clinicians in the diagnosis of benign and malignant spinal fracture on MRI, we applied the ResNet50 algorithm to develop a decision support system. METHODS: A total of 190 patients, 50 with malignant and 140 with benign fractures, were studied. The visual diagnosis was made by one senior MSK radiologist, one fourth-year resident, and one first-year resident. The MSK radiologist also gave the binary score for 15 qualitative imaging features. Deep learning was implemented using ResNet50, using one abnormal spinal segment selected from each patient as input. The T1W and T2W images of the lesion slice and its two neighboring slices were considered. The diagnostic performance was evaluated using tenfold cross-validation. RESULTS: The overall reading accuracy was 98, 96, and 66% for the senior MSK radiologist, fourth-year resident, and first-year resident, respectively. Of the 15 imaging features, 10 showed a significant difference between benign and malignant groups with p < = 0.001. The accuracy achieved by using the ResNet50 deep learning model for the identified abnormal vertebral segment was 92%. Compared to the first-year resident's reading, the model improved the sensitivity from 78 to 94% (p < 0.001) and the specificity from 61 to 91% (p < 0.001). CONCLUSION: Our deep learning-based model may provide information to assist less experienced clinicians in the diagnosis of spinal fractures on MRI. Other findings away from the vertebral body need to be considered to improve the model, and further investigation is required to generalize our findings to real-world settings.

Automatic Detection and Segmentation of Breast Cancer on MRI Using Mask R-CNN Trained on Non-Fat-Sat Images and Tested on Fat-Sat Images.

RATIONALE AND OBJECTIVES: Computer-aided methods have been widely applied to diagnose lesions on breast magnetic resonance imaging (MRI). The first step was to identify abnormal areas. A deep learning Mask Regional Convolutional Neural Network (R-CNN) was implemented to search the entire set of images and detect suspicious lesions. MATERIALS AND METHODS: Two DCE-MRI datasets were used, 241 patients acquired using non-fat-sat sequence for training, and 98 patients acquired using fat-sat sequence for testing. All patients have confirmed unilateral mass cancers. The tumor was segmented using fuzzy c-means clustering algorithm to serve as the ground truth. Mask R-CNN was implemented with ResNet-101 as the backbone. The neural network output the bounding boxes and the segmented tumor for evaluation using the Dice Similarity Coefficient (DSC). The detection performance, and the trade-off between sensitivity and specificity, was analyzed using free response receiver operating characteristic. RESULTS: When the precontrast and subtraction image of both breasts were used as input, the false positive from the heart and normal parenchymal enhancements could be minimized. The training set had 1469 positive slices (containing lesion) and 9135 negative slices. In 10-fold cross-validation, the mean accuracy = 0.86 and DSC = 0.82. The testing dataset had 1568 positive and 7264 negative slices, with accuracy = 0.75 and DSC = 0.79. When the obtained per-slice results were combined, 240 of 241 (99.5%) lesions in the training and 98 of 98 (100%) lesions in the testing datasets were identified. CONCLUSION: Deep learning using Mask R-CNN provided a feasible method to search breast MRI, localize, and segment lesions. This may be integrated with other artificial intelligence algorithms to develop a fully automatic breast MRI diagnostic system.

Solid tumor size for prediction of recurrence in large and giant non-functioning pituitary adenomas.

A subset of large non-functioning pituitary adenomas (lNFPA) and giant non-functioning pituitary adenomas (gNFPA) undergoes early progression/recurrence (P/R) after surgery. This study revealed the clinical and image predictors of P/R in lNFPA and gNFPA, with emphasis on solid tumor size. This retrospective study investigated the preoperative MR imaging features for the prediction of P/R in lNFPA (> 3 cm) and gNFPA (> 4 cm). Only the patients with a complete preoperative brain MRI and undergone postoperative MRI follow-ups for more than 1 year were included. From November 2010 to December 2020, a total of 34 patients diagnosed with lNFPA and gNFPA were included (median follow-up time 47.6 months) in this study. A total of twenty-three (23/34, 67.6%) patients had P/R, and the median time to P/R is 25.2 months. Solid tumor diameter (STD), solid tumor volume (STV), and extent of resection are associated with P/R (p < 0.05). Multivariate analysis showed large STV is a risk factor for P/R (p < 0.05) with a hazard ratio of 30.79. The cutoff points of STD and STV for prediction of P/R are 26 mm and 7.6 cm3, with AUCs of 0.78 and 0.79 respectively. Kaplan-Meier analysis of tumor P/R trends showed that patients with larger STD and STV exhibited shorter progression-free survival (p < 0.05). For lNFPA and gNFPA, preoperative STD and STV are significant predictors of P/R. The results offer objective and valuable information for treatment planning in this subgroup.

Diagnosis of Breast Cancer Using Radiomics Models Built Based on Dynamic Contrast Enhanced MRI Combined With Mammography.

OBJECTIVE: To build radiomics models using features extracted from DCE-MRI and mammography for diagnosis of breast cancer. MATERIALS AND METHODS: 266 patients receiving MRI and mammography, who had well-enhanced lesions on MRI and histologically confirmed diagnosis were analyzed. Training dataset had 146 malignant and 56 benign, and testing dataset had 48 malignant and 18 benign lesions. Fuzzy-C-means clustering algorithm was used to segment the enhanced lesion on subtraction MRI maps. Two radiologists manually outlined the corresponding lesion on mammography by consensus, with the guidance of MRI maximum intensity projection. Features were extracted using PyRadiomics from three DCE-MRI parametric maps, and from the lesion and a 2-cm bandshell margin on mammography. The support vector machine (SVM) was applied for feature selection and model building, using 5 datasets: DCE-MRI, mammography lesion-ROI, mammography margin-ROI, mammography lesion+margin, and all combined. RESULTS: In the training dataset evaluated using 10-fold cross-validation, the diagnostic accuracy of the individual model was 83.2% for DCE-MRI, 75.7% for mammography lesion, 64.4% for mammography margin, and 77.2% for lesion+margin. When all features were combined, the accuracy was improved to 89.6%. By adding mammography features to MRI, the specificity was significantly improved from 69.6% (39/56) to 82.1% (46/56), p<0.01. When the developed models were applied to the independent testing dataset, the accuracy was 78.8% for DCE-MRI and 83.3% for combined MRI+Mammography. CONCLUSION: The radiomics model built from the combined MRI and mammography has the potential to provide a machine learning-based diagnostic tool and decrease the false positive diagnosis of contrast-enhanced benign lesions on MRI.

BI-RADS Reading of Non-Mass Lesions on DCE-MRI and Differential Diagnosis Performed by Radiomics and Deep Learning.

BACKGROUND: A wide variety of benign and malignant processes can manifest as non-mass enhancement (NME) in breast MRI. Compared to mass lesions, there are no distinct features that can be used for differential diagnosis. The purpose is to use the BI-RADS descriptors and models developed using radiomics and deep learning to distinguish benign from malignant NME lesions. MATERIALS AND METHODS: A total of 150 patients with 104 malignant and 46 benign NME were analyzed. Three radiologists performed reading for morphological distribution and internal enhancement using the 5th BI-RADS lexicon. For each case, the 3D tumor mask was generated using Fuzzy-C-Means segmentation. Three DCE parametric maps related to wash-in, maximum, and wash-out were generated, and PyRadiomics was applied to extract features. The radiomics model was built using five machine learning algorithms. ResNet50 was implemented using three parametric maps as input. Approximately 70% of earlier cases were used for training, and 30% of later cases were held out for testing. RESULTS: The diagnostic BI-RADS in the original MRI report showed that 104/104 malignant and 36/46 benign lesions had a BI-RADS score of 4A-5. For category reading, the kappa coefficient was 0.83 for morphological distribution (excellent) and 0.52 for internal enhancement (moderate). Segmental and Regional distribution were the most prominent for the malignant group, and focal distribution for the benign group. Eight radiomics features were selected by support vector machine (SVM). Among the five machine learning algorithms, SVM yielded the highest accuracy of 80.4% in training and 77.5% in testing datasets. ResNet50 had a better diagnostic performance, 91.5% in training and 83.3% in testing datasets. CONCLUSION: Diagnosis of NME was challenging, and the BI-RADS scores and descriptors showed a substantial overlap. Radiomics and deep learning may provide a useful CAD tool to aid in diagnosis.

Using deep learning convolutional neural networks to automatically perform cerebral aqueduct CSF flow analysis.

Since the development of phase-contrast magnetic resonance imaging (PC-MRI), quantification of cerebrospinal fluid (CSF) flow across the cerebral aqueduct has been utilized for diagnosis of conditions such as normal pressure hydrocephalus (NPH). This study aims to develop an automated method of aqueduct CSF flow analysis using convolution neural networks (CNNs), which can replace the current standard involving manual segmentation of aqueduct region of interest (ROI). Retrospective analysis was performed on 333 patients who underwent PC-MRI, totaling 353 imaging studies. Aqueduct flow measurements using manual ROI placement was performed independently by two radiologists. Two types of CNNs, MultiResUNet and UNet, were trained using ROI data from the senior radiologist, with PC-MRI studies being randomly divided into training (80%) and validation (20%) datasets. Segmentation performance was assessed using Dice similarity coefficient (DSC), and CSF flow parameters were calculated from both manual and CNN-derived ROIs. MultiResUNet, UNet and second radiologist (Rater 2) had DSCs of 0.933, 0.928, and 0.867, respectively, with p < 0.001 between CNNs and Rater 2. Comparison of CSF flow parameters showed excellent intraclass correlation coefficients (ICCs) for MultiResUNet, with lowest correlation being 0.67. For UNet, lower ICCs of -0.01 to 0.56 were observed. Only 3/353 (0.8%) studies failed to have appropriate ROIs placed by MultiResUNet, compared to 12/353 (3.4%) failed cases for UNet. In conclusion, CNNs were able to measure aqueductal CSF flow with similar performance to a senior neuroradiologist. MultiResUNet demonstrated fewer cases of segmentation failure and more consistent flow measurements compared to the widely adopted UNet.