Cancer Imaging最新文献

Background: Accurate segmentation of pelvic and sacral tumors (PSTs) in multi-sequence magnetic resonance imaging (MRI) is essential for effective treatment and surgical planning.

Purpose: To develop a deep learning (DL) framework for efficient segmentation of PSTs from multi-sequence MRI.

Materials and methods: This study included a total of 616 patients with pathologically confirmed PSTs between April 2011 to May 2022. We proposed a practical DL framework that integrates a 2.5D U-net and MobileNetV2 for automatic PST segmentation with a fast annotation strategy across multiple MRI sequences, including T1-weighted (T1-w), T2-weighted (T2-w), diffusion-weighted imaging (DWI), and contrast-enhanced T1-weighted (CET1-w). Two distinct models, the All-sequence segmentation model and the T2-fusion segmentation model, were developed. During the implementation of our DL models, all regions of interest (ROIs) in the training set were coarse labeled, and ROIs in the test set were fine labeled. Dice score and intersection over union (IoU) were used to evaluate model performance.

Results: The 2.5D MobileNetV2 architecture demonstrated improved segmentation performance compared to 2D and 3D U-Net models, with a Dice score of 0.741 and an IoU of 0.615. The All-sequence model, which was trained using a fusion of four MRI sequences (T1-w, CET1-w, T2-w, and DWI), exhibited superior performance with Dice scores of 0.659 for T1-w, 0.763 for CET1-w, 0.819 for T2-w, and 0.723 for DWI as inputs. In contrast, the T2-fusion segmentation model, which used T2-w and CET1-w sequences as inputs, achieved a Dice score of 0.833 and an IoU value of 0.719.

Conclusions: In this study, we developed a practical DL framework for PST segmentation via multi-sequence MRI, which reduces the dependence on data annotation. These models offer solutions for various clinical scenarios and have significant potential for wide-ranging applications.

Background: To explore the value of intratumoral and peritumoral radiomics in preoperative prediction of anaplastic lymphoma kinase (ALK) mutation status and survival in patients with lung adenocarcinoma.

Methods: We retrospectively collected data from 505 eligible patients with lung adenocarcinoma from four hospitals (training and external validation sets 1-3). The CT-based radiomics features were extracted separately from the gross tumor volume (GTV) and GTV incorporating peritumoral 3-, 6-, 9-, 12-, and 15-mm regions (GPTV₃, GPTV₆, GPTV₉, GPTV₁₂, and GPTV₁₅), and screened the most relevant features to construct radiomics models to predict ALK (+). The combined model incorporated radiomics scores (Rad-scores) of the best radiomics model and clinical predictors was constructed. Performance was evaluated using receiver operating characteristic (ROC) analysis. Progression-free survival (PFS) outcomes were examined using the Cox proportional hazards model.

Results: In the four sets, 21.19% (107/505) patients were ALK (+). The GPTV₃ radiomics model using a support vector machine algorithm achieved the best predictive performance, with the highest average AUC of 0.811 in the validation sets. Clinical TNM stage and pleural indentation were independent predictors. The combined model incorporating the GPTV₃-Rad-score and clinical predictors achieved higher performance than the clinical model alone in predicting ALK (+) in three validation sets [AUC: 0.855 (95% CI: 0.766-0.919) vs. 0.648 (95% CI: 0.543-0.745), P = 0.001; 0.882 (95% CI: 0.801-0.962) vs. 0.634 (95% CI: 0.548-0.714), P < 0.001; 0.810 (95% CI: 0.727-0.877) vs. 0.663 (95% CI: 0.570-0.748), P = 0.006]. The prediction score of the combined model could stratify PFS outcomes in patients receiving ALK-TKI therapy (HR: 0.37; 95% CI: 0.15-0.89; P = 0.026) and immunotherapy (HR: 2.49; 95% CI: 1.22-5.08; P = 0.012).

Conclusion: The presented combined model based on GPTV₃ effectively mined tumor features to predict ALK mutation status and stratify PFS outcomes in patients with lung adenocarcinoma.

Purpose: This study aimed to select robust features against lung motion in a phantom study and use them as input to feature selection algorithms and machine learning classifiers in a clinical study to predict the lymphovascular invasion (LVI) of non-small cell lung cancer (NSCLC). The results of robust features were also compared with conventional techniques without considering the robustness of radiomic features.

Methods: An in-house developed lung phantom was developed with two 22mm lesion sizes based on a clinical study. A specific motor was built to simulate motion in two orthogonal directions. Lesions of both clinical and phantom studies were segmented using a Fuzzy C-means-based segmentation algorithm. After inducing motion and extracting 105 radiomic features in 4 feature sets, including shape, first-, second-, and higher-order statistics features from each region of interest (ROI) of the phantom image, statistical analyses were performed to select robust features against motion. Subsequently, these robust features and a total of 105 radiomic features were extracted from 126 clinical data. Various feature selection (FS) and multiple machine learning (ML) classifiers were implemented to predict the LVI of NSCLC, followed by comparing the results of predicting LVI using robust features with common conventional techniques not considering the robustness of radiomic features.

Results: Our results demonstrated that selecting robust features as input to FS algorithms and ML classifiers surges the sensitivity, which has a gentle negative effect on the accuracy and the area under the curve (AUC) of predictions compared with commonly used methods in 12 of 15 outcomes. The top performance of the LVI prediction was achieved by the NB classifier and RFE FS without considering the robustness of radiomic features with 95% area under the curve of AUC, 67% accuracy, and 100% sensitivity. Moreover, the top performance of the LVI prediction using robust features belonged to the NB classifier and Boruta feature selection with 92% AUC, 86% accuracy, and 100% sensitivity.

Conclusion: Robustness over various influential factors is critical and should be considered in a radiomic study. Selecting robust features is a solution to overcome the low reproducibility of radiomic features. Although setting robust features against motion in a phantom study has a minor negative impact on the accuracy and AUC of LVI prediction, it boosts the sensitivity of prediction to a large extent.

Purpose: To investigate change in diffusion weighted imaging (DWI) between pre-treatment (pre-) and after induction chemotherapy (post-IC) for long-term outcome prediction in advanced nasopharyngeal carcinoma (adNPC).

Materials and methods: Mean apparent diffusion coefficients (ADCs) of two DWIs (ADC_pre and ADC_post-IC) and changes in ADC between two scans (ΔADC%) were calculated from 64 eligible patients with adNPC and correlated with disease free survival (DFS), locoregional recurrence free survival (LRRFS), distant metastases free survival (DMFS), and overall survival (OS) using Cox regression analysis. C-indexes of the independent parameters for outcome were compared with that of RECIST response groups. Survival rates between two patient groups were evaluated and compared.

Results: Univariable analysis showed that high ΔADC% predicted good DFS, LRRFS, and DMFS p < 0.05), but did not predict OS (p = 0.40). Neither ADC_pre nor ADC_post-IC (p = 0.07 to 0.97) predicted outcome. Multivariate analysis showed that ΔADC% independently predicted DFS, LRRFS, and DMFS (p < 0.01 to 0.03). Compared with the RECIST groups, the ΔADC% groups (threshold of 34.2%) showed a higher c-index for 3-year (0.47 vs. 0.71, p < 0.01) and 5-year DFS (0.51 vs. 0.72, p < 0.01). Compared with patients with ΔADC%<34.2%, patients with ΔADC%≥34.2% had higher 3-year DFS, LRRFS and DMFS of 100%, 100% and 100%, respectively (p < 0.05).

Conclusion: Results suggest that ΔADC% was an independent predictor for long-term outcome and was superior to RECIST guideline for outcome prediction in adNPC. A ΔADC% threshold of ≥ 34.2% may be valuable for selecting patients who respond to treatment for de-escalation of treatment or post-treatment surveillance.

Background: Programmed death ligand 1 (PD-L1) expression status, closely related to immunotherapy outcomes, is a reliable biomarker for screening patients who may benefit from immunotherapy. Here, we developed and validated an interpretable machine learning (ML) model based on contrast-enhanced computed tomography (CECT) radiomics for preoperatively predicting PD-L1 expression status in patients with gastric cancer (GC).

Methods: We retrospectively recruited 285 GC patients who underwent CECT and PD-L1 detection from two medical centers. A PD-L1 combined positive score (CPS) of ≥ 5 was considered to indicate a high PD-L1 expression status. Patients from center 1 were divided into training (n = 143) and validation sets (n = 62), and patients from center 2 were considered a test set (n = 80). Radiomics features were extracted from venous-phase CT images. After feature reduction and selection, 11 ML algorithms were employed to develop predictive models, and their performance in predicting PD-L1 expression status was evaluated using areas under receiver operating characteristic curves (AUCs). SHapley Additive exPlanations (SHAP) were used to interpret the optimal model and visualize the decision-making process for a single individual.

Results: Nine features significantly associated with PD-L1 expression status were ultimately selected to construct the predictive model. The light gradient-boosting machine (LGBM) model demonstrated the best performance for PD-L1 high expression status prediction in the training, validation, and test sets, with AUCs of 0.841(95% CI: 0.773, 0.908), 0.834 (95% CI:0.729, 0.939), and 0.822 (95% CI: 0.718, 0.926), respectively. The SHAP summary and bar plots illustrated that a feature's value affected the feature's impact attributed to the model. The SHAP waterfall plots were used to visualize the decision-making process for a single individual.

Conclusion: Our CT radiomics-based LGBM model may aid in preoperatively predicting PD-L1 expression status in GC patients, and the SHAP method may improve the interpretability of this model.

Background: Despite the increasing interest in abbreviated protocols, we adopted an extended protocol for all prostate MRIs. In this study, we assessed the benefits of an extended prostate MRI protocol, measured by the number and the clinical importance of additional findings (AFs) and their impact on patient management.

Methods: In a single-center study, we retrospectively included 1282 patients undergoing prostate MRI between 01.10.2018 and 30.04.2022. Additional findings were defined as any pathology not located in the prostate or the seminal vesicles. These were classified as related or unrelated to prostate cancer (PCa). The latter were divided into groups based on low, moderate, or high clinical significance (group 1, 2, and 3). A finding unrelated to PCa was judged to be clinically significant (group 2: moderate, group 3: high) if further diagnostic investigations, or treatment was necessary. The degree of urgency of the latter determined moderate and high significance. For group 3 findings, a change in management was defined as further workup.

Results: A total of 5206 AFs was recorded in 1240/1282 patients. One hundred and twenty-three (2.4% of all findings) extra-prostatic PCa related AFs were found in 106 (8.3% of all patients) patients. The remaining 5083 (97.6% of all findings) findings were not related to PCa, of which 3155 (60.6%), 1770 (34.0%), and 158 (3.0%) were assigned to groups 1, 2, and 3, respectively. A management shift was identified in 49 (3.8% of all patients) patients of group 3.

Conclusion: The extended prostate MRI protocol shows a considerable prevalence of AFs of which more than a third are clinically significant, related or unrelated to PCa (groups 2 and 3). A substantial percentage (8.3%) of patients have extra-prostatic PCa-related AFs that change the patient's disease stage and management. However, a change in management due to AFs unrelated to PCA that belong to group 3 is observed in less than 4% of all patients. The choice between extended and abbreviated prostate MRI protocols should be made based on available resources.

Background: Non-small cell lung cancer (NSCLC) is a common neoplasm with poor prognosis in advanced stages. The clinical work-up in patients with locally advanced NSCLC mostly includes ¹⁸F-fluorodeoxyglucose positron emission tomography computed tomography (¹⁸F-FDG PET/CT) because of its high sensitivity for malignant lesion detection; however, specificity is lower. Diverging results exist whether whole-body MRI (WB-MRI) improves the staging accuracy in advanced lung cancer. Considering WB-MRI being a more time-consuming examination compared to brain MRI, it is important to establish whether or not additional value is found in detecting and characterizing malignant lesions. The purpose of this study is to investigate the value of additional whole-body magnetic resonance imaging, instead of only brain MRI, together with ¹⁸F-FDG PET/CT in staging patients with advanced NSCLC planned for curative treatment.

Material and methods: In a prospective single center study, 28 patients with NSCLC stage 3 or oligometastatic disease were enrolled. In addition to ¹⁸F-FDG PET/CT, they underwent WB-MRI including the thorax, abdomen, spine, pelvis, and contrast-enhanced examination of the brain and liver. ¹⁸F-FDG PET/CT and WB-MRI were separately evaluated by two blinded readers, followed by consensus reading in which the likelihood of malignancy was assessed in detected lesions. Imaging and clinical follow-up for at least 12 months was used as reference standard. Statistical analyses included Fischer's exact test and Clopped-Pearson.

Results: 28 patients (mean age ± SD 70.5 ± 8.4 years, 19 women) were enrolled. WB-MRI and FDG-PET/CT both showed maximum sensitivity and specificity for primary tumor diagnosis and similar sensitivity (p = 1.00) and specificity (p = 0.70) for detection of distant metastases. For diagnosis of lymph node metastases, WB-MRI showed lower sensitivity, 0.65 (95% CI: 0.38-0.86) than FDG-PET/CT, 1.00 (95% CI: 0.80-1.00) (p < 0.05), but similar specificity (p = 0.59).

Conclusions: WB-MRI in conjunction with ¹⁸F-FDG PET/CT provides no additional value over MRI of the brain only, in staging patients with advanced NSCLC.

Trial registration: Registered locally and approved by the Uppsala University Hospital committee, registration number ASMR020.

Background: In 2020, we introduced the Greater Omentum Imaging-Reporting and Data System (GOI-RADS), a novel classification system related to peritoneal lesions. However, its clinical application remained unvalidated.

Objective: This study aimed to validate GOI-RADS, optimize its parameters for a new grading system, and explore its clinical usefulness.

Methods: A retrospective-prospective study was conducted to validate and refine the GOI-RADS system. The study consisted of two phases: a retrospective validation phase and a prospective application phase. The first phase included patients with peritoneal lesions from 2019 to 2021, classified by GOI-RADS and verified against pathology. Contrast-enhanced ultrasound (CEUS) and real-time elastography (RTE) data were collected for developing a new grading system. Odds ratios optimized parameters. The second phase (2021-2024) assessed diagnostic consistency among sonographers and performance of grading systems.

Results: Among 215 patients with peritoneal lesions, the actual malignancy rates for GOI-RADS 2 (40.00%) and GOI-RADS 3 (61.22%) were much higher than predicted (5.56% and 37.25%). Combining CEUS and RTE parameters showed varying sensitivity and specificity: RTE + GOI-RADS (95.35%, 55.56%) and CEUS + GOI-RADS (96.51%, 44.44%). However, the grading system based on multiple ultrasound parameters, specifically when incorporating RTE, CEUS parameters, and GOI-RADS (Multi-GOIRADS), exhibited the highest diagnostic sensitivity and specificity of 88.37% and 83.33%, respectively. Its simplified version, sMulti-GOIRADS, had sensitivity of 73.26% and specificity of 94.44%. In the prospective study involving three sonographers of different qualifications, the use of sMulti-GOIRADS was found to be the most time-efficient and showed excellent diagnostic consistency among them. In contrast, Multi-GOIRADS required more time for scoring but offered superior diagnostic performance, particularly among senior sonographers (88.35% and 91.43%).

Conclusions: This study proposes a multiparametric ultrasound-based imaging-reporting and data system for risk stratification of omental malignancy, Multi-GOIRADS, and presents an optimized and simplified version, sMulti-GOIRADS, which demonstrates excellent diagnostic consistency and performance in clinical applications.

Background: To construct and assess a deep learning (DL) signature that employs computed tomography imaging to predict the expression status of programmed cell death ligand 1 in patients with bladder cancer (BCa).

Methods: This retrospective study included 190 patients from two hospitals who underwent surgical removal of BCa (training set/external validation set, 127/63). We used convolutional neural network and radiomics machine learning technology to generate prediction models. We then compared the performance of the DL signature with the radiomics machine learning signature and selected the optimal signature to build a nomogram with the clinical model. Finally, the internal forecasting process of the DL signature was explained using Shapley additive explanation technology.

Results: On the external validation set, the DL signature had an area under the curve of 0.857 (95% confidence interval: 0.745-0.932), and demonstrated superior prediction performance in comparison with the other models. SHAP expression images revealed that the prediction of PD-L1 expression status is mainly influenced by the tumor edge region, particularly the area close to the bladder wall.

Conclusions: The DL signature performed well in comparison with other models and proved to be a valuable, dependable, and interpretable tool for predicting programmed cell death ligand 1 expression status in patients with BCa.

Objectives: To evaluate the feasibility and value of dynamic contrast-enhanced MRI (DCE-MRI) quantitative analysis and MRI-based radiomics in predicting the efficacy of microwave ablation (MWA) in lung cancers (LCs).

Methods: Forty-three patients with LCs who underwent DCE-MRI within 24 h of receiving MWA were enrolled in the study and divided into two groups according to the modified response evaluation criteria in solid tumors (m-RECIST) criteria: the effective treatment (complete response + partial response + stable disease, n = 28) and the ineffective treatment (progressive disease, n = 15). DCE-MRI datasets were processed by Omni. Kinetics software, using the extended tofts model (ETM) and exchange model (ECM) to yield pharmacokinetic parameters and their histogram features. Changes in quantitative perfusion parameters were compared between the two groups. Scientific research platform ( https://medresearch.shukun.net/ ) was used for radiomics analysis. A total of 1874 radiomic features were extracted for each tumor by manually segmentation of T1WI and Contrast-enhanced of T1WI (Ce-T1WI) fat inhibition sequence. The performances of radiomics models were evaluated by the receiver operating characteristic curve. Based on radiomics features, survival curves were generated by Kaplan-Meier survival analysis to evaluate patient outcomes. P < 0.05 was set for the significance threshold.

Results: The V_p value of ECM was significantly higher in the ineffective group compared to the effective group (p = 0.027). Additionally, the skewness, and kurtosis of V_p (p = 0.020 and 0.013, respectively) derived from ETM and F_p (p = 0.027 and 0.030, respectively) from ECM as well as the quantiles were higher in the ineffective group than in the effective group. Significant statistical differences were observed in the energy and entropy of V_e (p = 0.044 and 0.025, respectively) and V_p (p = 0.025 and 0.026, respectively) between the two groups. In the process of model construction, seven features from T1WI, five features from Ce-T1WI, and ten features from combined sequences were ultimately selected. The area under the curve (AUC) values for the T1WI model, Ce-T1WI model, and combined model were 0.910, 0.890, 0.965 in the training group, and 0.850, 0.700, 0.850 in the test group, respectively.

Conclusions: DCE-MRI quantitative analysis and MRI-based radiomics may be helpful in assessing the early response to MWA in patients with LCs.