Academic Radiology最新文献

Rationale and objectives: To develop and evaluate an AI algorithm that detects breast cancer in MRI scans up to one year before radiologists typically identify it, potentially enhancing early detection in high-risk women.

Materials and methods: A convolutional neural network (CNN) AI model, pre-trained on breast MRI data, was fine-tuned using a retrospective dataset of 3029 MRI scans from 910 patients. These contained 115 cancers that were diagnosed within one year of a negative MRI. The model aimed to identify these cancers, with the goal of predicting cancer development up to one year in advance. The network was fine-tuned and tested with 10-fold cross-validation. Mean age of patients was 52 years (range, 18-88 years), with average follow-up of 4.3 years (range 1-12 years).

Results: The AI detected cancers one year earlier with an area under the ROC curve of 0.72 (0.67-0.76). Retrospective analysis by a radiologist of the top 10% highest risk MRIs as ranked by the AI could have increased early detection by up to 30%. (35/115, CI:22.2-39.7%, 30% sensitivity). A radiologist identified a visual correlate to biopsy-proven cancers in 83 of prior-year MRIs (83/115, CI: 62.1-79.4%). The AI algorithm identified the anatomic region where cancer would be detected in 66 cases (66/115, CI:47.8-66.5%); with both agreeing in 54 cases (54/115, CI:%37.5-56.4%).

Conclusion: This novel AI-aided re-evaluation of "benign" breasts shows promise for improving early breast cancer detection with MRI. As datasets grow and image quality improves, this approach is expected to become even more impactful.

Rationale and objectives: This study aimed to address the challenge of predicting treatment outcomes for patients with uterine fibroids undergoing high-intensity focused ultrasound (HIFU) ablation. We developed medical-assisted diagnostic models to accurately predict the ablation rates and volume reduction rates, thus assessing both short-term and long-term treatment effects of fibroids.

Materials and methods: For the ablation rate prediction, our study included 348 fibroids, categorized into 181 fully ablated and 167 inadequately ablated fibroids. Using multimodal MRI sequences and clinical characteristics, coupled with data preprocessing steps such as feature extraction, testing, and screening, we constructed an ensemble model for predicting preoperative ablation rates. In the volume reduction rate study, we analyzed 253 fibroids, divided into 142 high-volume responders and 111 low-volume responders. Based on clinical characteristics and T2-weighted image (T2WI) sequences, along with lesion delineation, feature normalization, and other preprocessing steps, we developed an inter-slice information fusion model for predicting preoperative volume reduction rates.

Results: The ensemble model demonstrated an accuracy of 0.800 and an area under the curve (AUC) of 0.830 on the test set, while the inter-slice information fusion model achieved an accuracy of 0.808 and an AUC of 0.891. Both models showed superior predictive performance compared to existing models.

Conclusion: The ensemble and inter-slice information fusion models developed in this study exhibit robust predictive capabilities, offering valuable support for clinicians in selecting patients for HIFU treatment. These models hold potential for enhancing patient outcomes through tailored treatment planning.

Rationale and objectives: Artificial intelligence (AI) algorithms in radiology capable of detecting urgent findings have gained significant traction in recent years, but the impact of these algorithms on real-world clinical practice remains unclear with need for scientific investigation. Our study investigates the diagnostic accuracy and impact on radiologist report turnaround times of an FDA-approved AI tool for pneumothorax (PTx) detection on inpatient chest X-rays (CXR) in our institution's radiology practice at a large academic medical center.

Materials and methods: This retrospective study included 27,397 frontal, single-view CXRs of adult inpatients collected consecutively between August 2020 and April 2021 following deployment of an AI-based PTx detection and picture archiving and communication system (PACS) alert system. 12,728 CXRs were acquired within the AI-integrated system while 14,669 CXRs were acquired outside of the system. Receiver operator characteristic (ROC) analysis was conducted with final radiology reports as the reference standard to evaluate diagnostic accuracy of the AI algorithm in detection of PTx. Wilcoxon rank sum tests were conducted to evaluate the effect of the AI-integrated alert system on radiologist reporting times.

Results: Area under ROC curve (AUC) for the AI tool was.78 with sensitivity of .60 and specificity of .97. When selecting for moderate/large PTx, AUC, sensitivity and specificity increased to .93, .89 and .96, respectively. Median reporting time in CXRs with radiologist-confirmed PTx was reduced by 46% in those with AI integration as compared to those without AI integration (100 vs. 186 min, p < .001).

Conclusion: Real-world deployment of an AI-integrated system capable of detecting PTx and generating alerts within PACS achieved a strong AUC for clinically actionable PTx (i.e., moderate- or large-sized) while substantially reducing median radiologist reporting times, enabling swifter clinical response to a critical but treatable condition.

Objective: We investigated the value of PET/CT-based multimodal parameters in predicting the degree of differentiation and epidermal growth factor receptor (EGFR) mutations in invasive lung adenocarcinoma (ILA) and assessed the correlation between PET/CT-based multimodal parameters and Ki67.

Methods: We retrospectively collected 113 patients with ILA who underwent PET/CT examination, and differences in PET/CT multimodal parameters between different differentiation groups were analyzed. Binary logistic regression was used to establish a multiparameter model for predicting EGFR mutation, and ROC curve was used to compare the diagnostic efficiency. Independent predictors of the Ki67 index were screened using multiple linear regression analysis.

Results: The poorly differentiated group was more likely to have large-diameter, solid foci, pleural pulling signs, and vacuolar signs compared with other groups (all P < 0.05). The differences in metabolic tumor volume (MTV) and total lesion glycolysis (TLG) in all three different differentiated groups were statistically significant compared to the other parameters (all P < 0.05). The PET/CT regression model predicted EGFR mutations with an AUC of 0.820 and was higher than other models; the sensitivity, specificity, positive predictive value, and negative predictive value for discriminating EGFR mutations were 84.74%, 70.37%, 75.76%, and 80.85%, respectively. PET/CT multiple linear regression analysis showed that vascular convergence, SUVpeak, MTV, and TLG explaining 62.0% changes in Ki67 (R² = 0.620). SUVpeak, MTV, and TLG (r = 0.580, r = 0.662, and r = 0.680, all P < 0.001) were all strongly correlated with increased Ki67 index.

Conclusion: MTV and TLG can better identify the degree of ILA differentiation compared to CT and other PET parameters. The nomogram constructed by multimodal PET/CT parameters can better dynamically monitor the changes of EGFR status and Ki67 index.

Rationale and objectives: Accurate prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC) is crucial for guiding treatment. This study evaluates and compares the performance of clinicoradiologic, traditional radiomics, deep-learning radiomics, feature fusion, and decision fusion models based on multi-region MR habitat imaging using six machine-learning classifiers.

Materials and methods: We retrospectively included 300 HCC patients. The intratumoral and peritumoral regions were segmented into distinct habitats, from which radiomics and deep-learning features were extracted using arterial phase MR images. To reduce feature dimensionality, we applied intra-class correlation coefficient (ICC) analysis, Pearson correlation coefficient (PCC) filtering, and recursive feature elimination (RFE). Based on the selected optimal features, prediction models were constructed using decision tree (DT), K-nearest neighbors (KNN), logistic regression (LR), random forest (RF), support vector machine (SVM), and XGBoost (XGB) classifiers. Additionally, fusion models were developed utilizing both feature fusion and decision fusion strategies. The performance of these models was validated using the area under the receiver operating characteristic curve (ROC AUC), calibration curves, and decision curve analysis.

Results: The decision fusion model (VOI-Peri10-1) using LR and integrating clinicoradiologic, radiomics, and deep-learning features achieved the highest AUC of 0.808 (95% confidence interval [CI]: 0.807-0.912) in the test cohort, with good calibration (Hosmer-Lemeshow test, P > 0.050) and clinical net benefit.

Conclusion: The LR-based decision fusion model integrating clinicoradiologic, radiomics, and deep-learning features shows promise for preoperative prediction of MVI in HCC, aiding in patient outcome predictions and personalized treatment planning.

Rationale and objectives: Hemorrhagic transformation (HT) is one of the most serious complications in patients with acute ischemic stroke (AIS) following reperfusion therapy. The purpose of this study is to develop and validate deep learning (DL) models utilizing multiphase computed tomography angiography (CTA) and computed tomography perfusion (CTP) images for the fully automated prediction of HT.

Materials and methods: In this multicenter retrospective study, a total of 229 AIS patients who underwent reperfusion therapy from June 2019 to May 2022 were reviewed. Data set 1, comprising 183 patients from two hospitals, was utilized for training, tuning, and internal validation. Data set 2, consisting of 46 patients from a third hospital, was employed for external testing. DL models were trained to extract valuable information from multiphase CTA and CTP images. The DenseNet architecture was used to construct the DL models. We developed single-phase, single-parameter models, and combined models to predict HT. The models were evaluated using receiver operating characteristic curves.

Results: Sixty-nine (30.1%) of 229 patients (mean age, 66.9 years ± 10.3; male, 144 [66.9%]) developed HT. Among the single-phase models, the arteriovenous phase model demonstrated the highest performance. For single-parameter models, the time-to-peak model was superior. When considering combined models, the CTA-CTP model provided the highest predictive accuracy.

Conclusions: DL models for predicting HT based on multiphase CTA and CTP images can be established and performed well, providing a reliable tool for clinicians to make treatment decisions.

Rationale and objectives: Misregistration artifacts between the PET and attenuation correction CT (CTAC) exams can degrade image quality and cause diagnostic errors. Deep learning (DL)-warped elastic registration methods have been proposed to improve misregistration errors.

Materials and methods: 30 patients undergoing routine oncologic examination (20 ¹⁸F-FDG PET/CT and 10 ⁶⁴Cu-DOTATATE PET/CT) were retrospectively identified and compared using unmodified CTAC, and a DL-augmented spatial transformation CT attenuation map. Primary endpoints included differences in subjective image quality and standardized uptake values (SUV). Exams were randomized to reduce reader bias, and three radiologists rated image quality across six anatomic sites using a modified Likert scale. Measures of local bias and lesion SUV were also quantitatively evaluated.

Results: The DL attenuation correction methods were associated with higher image quality and reduced misregistration artifacts (Mean ¹⁸F-FDG quality rating=3.5-3.8 for DL vs 3.2-3.5 for standard reconstruction (STD); Mean ⁶⁴Cu-DOTATATE quality rating= 3.2-3.4 for DL vs 2.1-3.3; P < 0.05 for STD, for all except ⁶⁴Cu-DOTATATE inferior spleen). Percent change in superior liver SUV_mean for ¹⁸F-FDG and ⁶⁴Cu-DOTATATE were 5.3 ± 4.9 and 8.2 ± 4.1%, respectively. Measures of signal-to-noise ratio were significantly improved for the DL over STD (Hepatopulmonary index (HPI) [¹⁸F-FDG] = 4.5 ± 1.2 vs 4.0 ± 1.1, P < 0.001; HPI [⁶⁴Cu-DOTATATE] = 16.4 ± 16.9 vs 12.5 ± 5.5, P = 0.039).

Conclusion: Deep learning elastic registration for CT attenuation correction maps on routine oncology PET/CT decreases misregistration artifacts, with a greater impact on PET scans with longer acquisition times.

Rationale and objectives: Underrepresentation of minorities is a worsening issue in the field of radiology. Early educational interventions are a promising approach to mitigating this disparity. We present an approach for a radiology department to increase community outreach via establishment of an educational program for local public high school students and building a mentorship pipeline for radiology education.

Materials and methods: The department of radiology committee for Inclusion Diversity and Equity (IDE), in collaboration with the Office of Outreach, Education and Research (OER), invited yearly cohorts of 25 and 24 public high school students in 2022 and 2023, respectively, to an on-site educational event featuring rotating small group hands-on workshops in a multi-stage format. The event inspired students to consider various careers in radiology and their corresponding academic pathways after high school. Post-workshop surveys featuring Likert scale and open-ended questions were administered to collect student reflections and feedback. Analysis was conducted to assess student understanding, interest in radiological careers, and opportunities for future event improvements. Longitudinal mentorship was established between students and point-persons to provide continued career guidance.

Results: For two consecutive cohort years, the program received high scores on clarity of presentations and increased student awareness of opportunities within radiology. Standout positive elements included interactive sessions, hands-on activities, and the discovery of radiology as a collaborative field. Of the initial student group, one student went on to enroll in a radiography program. To date, five participants have returned for shadowing experiences, three of whom are currently enrolled in science undergraduate programs, including one pre-medical student.

Conclusion: We present an accessible and effective approach for a radiology department to collaboratively increase community outreach and improve minority representation through early educational programming and establishing a longitudinal pipeline mentorship program for public high school students.

Rationale and objectives: Microsatellite instability (MSI) stratification can guide the clinical management of patients with colorectal cancer (CRC). This study aimed to establish a radiomics model for predicting the MSI status of patients with CRC before treatment.

Materials and methods: This retrospective study was performed on 366 patients diagnosed with CRC who underwent preoperative magnetic resonance imaging (MRI) and immunohistochemical staining between February 2016 and September 2023. The participants were divided randomly into training and testing cohorts in a 7:3 ratio. The tumor volume of interest (VOI) was manually delineated on T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) sequences using 3D Slicer software, and radiomics features were extracted. Feature selection was performed using the least absolute shrinkage and selection operator method. A radiomics nomogram was developed using multiple logistic regression, and the predictive performance of the models was evaluated and compared using receiver operating characteristic curves. The calibration curve, clinical decision curve analysis (DCA) and clinical impact curve (CIC) were used to evaluate the clinical application value of the model.

Results: The radiomics normogram combined with history of chronic enteritis, tumor location, MR-reported inflammatory response, D2-40, carcinoembryonic antigen, tumor protein 53, and monocyte was an excellent predictive tool. The area under the curve for the training and testing cohorts were 0.927 and 0.984, respectively. The DCA and CIC demonstrated favorable clinical application and net benefit.

Conclusions: A radiomics nomogram based on T2WI and ADC sequences and clinicopathologic features can effectively and noninvasively predict the MSI status in CRC. This approach helps clinicians in stratifying CRC patients and making clinical decisions for personalized treatment.