Development and External Validation of Deep-Learning-Based Tumor Grading Models in Soft-Tissue Sarcoma Patients Using MR Imaging.
Ontology highlight
ABSTRACT: Background
In patients with soft-tissue sarcomas, tumor grading constitutes a decisive factor to determine the best treatment decision. Tumor grading is obtained by pathological work-up after focal biopsies. Deep learning (DL)-based imaging analysis may pose an alternative way to characterize STS tissue. In this work, we sought to non-invasively differentiate tumor grading into low-grade (G1) and high-grade (G2/G3) STS using DL techniques based on MR-imaging.Methods
Contrast-enhanced T1-weighted fat-saturated (T1FSGd) MRI sequences and fat-saturated T2-weighted (T2FS) sequences were collected from two independent retrospective cohorts (training: 148 patients, testing: 158 patients). Tumor grading was determined following the French Federation of Cancer Centers Sarcoma Group in pre-therapeutic biopsies. DL models were developed using transfer learning based on the DenseNet 161 architecture.Results
The T1FSGd and T2FS-based DL models achieved area under the receiver operator characteristic curve (AUC) values of 0.75 and 0.76 on the test cohort, respectively. T1FSGd achieved the best F1-score of all models (0.90). The T2FS-based DL model was able to significantly risk-stratify for overall survival. Attention maps revealed relevant features within the tumor volume and in border regions.Conclusions
MRI-based DL models are capable of predicting tumor grading with good reproducibility in external validation.
SUBMITTER: Navarro F
PROVIDER: S-EPMC8227009 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BackgroundPrognostic nomograms for patients with extremity soft tissue sarcoma (eSTS) typically predict survival or the occurrence of local recurrence or distant metastasis at time of surgery. Our aim was to develop and externally validate a dynamic prognostic nomogram for overall survival in eSTS survivors for use during follow-up.MethodsAll primary eSTS patients operated with curative intent between 1994 and 2013 at three European and one Canadian sarcoma centers formed the development cohort. Patients with Fédération Française des Centres de Lutte Contre le Cancer (FNCLCC) grade II and grade III eSTS operated between 2000 and 2016 at seven other European reference centers formed the external validation cohort. We used a landmark analysis approach and a multivariable Cox model to create a dynamic nomogram; the prediction window was fixed at five years. A backward procedure based on the Akaike Information Criterion was adopted for variable selection. We tested the nomogram performance in terms of calibration and discrimination.FindingsThe development and validation cohorts included 3740 and 893 patients, respectively. The variables selected applying the backward procedure were patient's age, tumor size and its interaction with landmark time, tumor FNCLCC grade and its interaction with landmark time, histology, and both local recurrence and distant metastasis (as first event) indicator variables. The nomogram showed good calibration and discrimination. Harrell C indexes at different landmark times were between 0.776 (0.761-0.790) and 0.845 (0.823-0.862) in the development series and between 0.675 (0.643-0.704) and 0.810 (0.775-0.844) in the validation series.InterpretationA new dynamic nomogram is available to predict 5-year overall survival at different times during the first three years of follow-up in patients operated for primary eSTS. This nomogram allows physicians to update the individual survival prediction during follow-up on the basis of baseline variables, time elapsed from surgery and first-event history.
| S-EPMC6933187 | biostudies-literature
Project description:Background: Accurate glioma grading before surgery is of the utmost importance in treatment planning and prognosis prediction. But previous studies on magnetic resonance imaging (MRI) images were not effective enough. According to the remarkable performance of convolutional neural network (CNN) in medical domain, we hypothesized that a deep learning algorithm can achieve high accuracy in distinguishing the World Health Organization (WHO) low grade and high grade gliomas. Methods: One hundred and thirteen glioma patients were retrospectively included. Tumor images were segmented with a rectangular region of interest (ROI), which contained about 80% of the tumor. Then, 20% data were randomly selected and leaved out at patient-level as test dataset. AlexNet and GoogLeNet were both trained from scratch and fine-tuned from models that pre-trained on the large scale natural image database, ImageNet, to magnetic resonance images. The classification task was evaluated with five-fold cross-validation (CV) on patient-level split. Results: The performance measures, including validation accuracy, test accuracy and test area under curve (AUC), averaged from five-fold CV of GoogLeNet which trained from scratch were 0.867, 0.909, and 0.939, respectively. With transfer learning and fine-tuning, better performances were obtained for both AlexNet and GoogLeNet, especially for AlexNet. Meanwhile, GoogLeNet performed better than AlexNet no matter trained from scratch or learned from pre-trained model. Conclusion: In conclusion, we demonstrated that the application of CNN, especially trained with transfer learning and fine-tuning, to preoperative glioma grading improves the performance, compared with either the performance of traditional machine learning method based on hand-crafted features, or even the CNNs trained from scratch.
| S-EPMC6250094 | biostudies-literature
Project description:Rationale: Hypoxic regions (habitats) within tumors are heterogeneously distributed and can be widely variant. Hypoxic habitats are generally pan-therapy resistant. For this reason, hypoxia-activated prodrugs (HAPs) have been developed to target these resistant volumes. The HAP evofosfamide (TH-302) has shown promise in preclinical and early clinical trials of sarcoma. However, in a phase III clinical trial of non-resectable soft tissue sarcomas, TH-302 did not improve survival in combination with doxorubicin (Dox), possibly due to a lack of patient stratification based on hypoxic status. Therefore, we used magnetic resonance imaging (MRI) to identify hypoxic habitats and non-invasively follow therapies response in sarcoma mouse models. Methods: We developed deep-learning (DL) models to identify hypoxia, using multiparametric MRI and co-registered histology, and monitored response to TH-302 in a patient-derived xenograft (PDX) of rhabdomyosarcoma and a syngeneic model of fibrosarcoma (radiation-induced fibrosarcoma, RIF-1). Results: A DL convolutional neural network showed strong correlations (>0.76) between the true hypoxia fraction in histology and the predicted hypoxia fraction in multiparametric MRI. TH-302 monotherapy or in combination with Dox delayed tumor growth and increased survival in the hypoxic PDX model (p<0.05), but not in the RIF-1 model, which had a lower volume of hypoxic habitats. Control studies showed that RIF-1 resistance was due to hypoxia and not other causes. Notably, PDX tumors developed resistance to TH-302 under prolonged treatment that was not due to a reduction in hypoxic volumes. Conclusion: Artificial intelligence analysis of pre-therapy MR images can predict hypoxia and subsequent response to HAPs. This approach can be used to monitor therapy response and adapt schedules to forestall the emergence of resistance.
| S-EPMC8039958 | biostudies-literature
Project description:Soft tissue sarcoma is an umbrella term for a group of rare cancers that are difficult to treat. In addition to surgery, neoadjuvant chemotherapy has shown the potential to downstage tumors and prevent micrometastases. However, finding effective therapeutic targets remains a research challenge. Here, a previously developed computational approach called mechanistic models of signaling pathways has been employed to unravel the impact of observed changes at the gene expression level on the ultimate functional behavior of cells. In the context of such a mechanistic model, RNA-Seq counts sourced from the Recount3 resource, from The Cancer Genome Atlas (TCGA) Sarcoma project, and non-diseased sarcomagenic tissues from the Genotype-Tissue Expression (GTEx) project were utilized to investigate signal transduction activity through signaling pathways. This approach provides a precise view of the relationship between sarcoma patient survival and the signaling landscape in tumors and their environment. Despite the distinct regulatory alterations observed in each sarcoma subtype, this study identified 13 signaling circuits, or elementary sub-pathways triggering specific cell functions, present across all subtypes, belonging to eight signaling pathways, which served as predictors for patient survival. Additionally, nine signaling circuits from five signaling pathways that highlighted the modifications tumor samples underwent in comparison to normal tissues were found. These results describe the protective role of the immune system, suggesting an anti-tumorigenic effect in the tumor microenvironment, in the process of tumor cell detachment and migration, or the dysregulation of ion homeostasis. Also, the analysis of signaling circuit intermediary proteins suggests multiple strategies for therapy.
| S-EPMC10572617 | biostudies-literature
Project description:ObjectivesTo build and evaluate a deep learning radiomics nomogram (DLRN) for preoperative prediction of lung metastasis (LM) status in patients with soft tissue sarcoma (STS).MethodsIn total, 242 patients with STS (training set, n=116; external validation set, n=126) who underwent magnetic resonance imaging were retrospectively enrolled in this study. We identified independent predictors for LM-status and evaluated their performance. The minimum redundancy maximum relevance (mRMR) method and least absolute shrinkage and selection operator (LASSO) algorithm were adopted to screen radiomics features. Logistic regression, decision tree, random forest, support vector machine (SVM), and adaptive boosting classifiers were compared for their ability to predict LM. To overcome the imbalanced distribution of the LM data, we retrained each machine-learning classifier using the synthetic minority over-sampling technique (SMOTE). A DLRN combining the independent clinical predictors with the best performing radiomics prediction signature (mRMR+LASSO+SVM+SMOTE) was established. Area under the receiver operating characteristics curve (AUC), calibration curves, and decision curve analysis (DCA) were used to assess the performance and clinical applicability of the models.ResultComparisons of the AUC values applied to the external validation set revealed that the DLRN model (AUC=0.833) showed better prediction performance than the clinical model (AUC=0.664) and radiomics model (AUC=0.799). The calibration curves indicated good calibration efficiency and the DCA showed the DLRN model to have greater clinical applicability than the other two models.ConclusionThe DLRN was shown to be an accurate and efficient tool for LM-status prediction in STS.
| S-EPMC9265249 | biostudies-literature
Project description:IntroductionSoft tissue sarcomas are a group of malignancies that commonly occur in the extremities. As deep lesions may exist within the confines of the muscular fascia, we postulate that local recurrence rates are higher for superficial soft tissue sarcomas managed by the standard of care.Materials and methodsA retrospective review was performed on 90 patients who underwent surgical resection of soft tissue sarcomas of the extremity from 2007 to 2015. Patients with minimum 2-year follow-up and adequate operative, pathologic, and clinical outcomes data were included.ResultsMean age was 54 ± 18 years with 49 (54.4%) patients being male. Lesions in 77.8% of cases were deep, and 22.2% were superficial to fascia. Following the index surgical resection, a total of 33 (36.7%) patients had positive margins. A total of 17 (18.9%) patients had a local recurrence. Overall, 3-year survival was 92.7%, and 5-year survival was 79.0%. Five-year recurrence-free survival of deep sarcomas was 91.1% versus 58.2% of superficial lesions (p = 0.006). Patients with higher tumor depth had lower odds of experiencing a local recurrence (HR 0.26 [95% CI 0.09-0.72]). Local recurence rates was also associated with positive surgical margins on initial resection (33.3% versus 12.3%) (p = 0.027).ConclusionsIn this series, superficial tumor depth was associated with local recurrence of soft tissue sarcomas of the extremity following surgical resection. Positive surgical margins was also associated with local recurrence.
| S-EPMC11319495 | biostudies-literature
Project description:As the most common transcriptional regulators, zinc finer proteins (ZNFs) play vital roles in occurrence and progression of malignant tumors. Whereas, information regarding the roles of ZNFs in soft tissue sarcomas (STS) remains scarce. In this study, a comprehensive bioinformatics analysis investigating roles of ZNFs in STS was performed. Initially, we extracted raw datasets of differentially expressed ZNFs from GSE2719. Using a sequence of bioinformatics methods, we then investigated the prognostic significance, function, and molecular subtype of these differentially expressed ZNFs. In addition, CCK8 and plate clone formation assays were used to explore the effect of ZNF141 on STS cells. A total of 110 differentially expressed ZNFs were identified. Nine ZNFs (HLTF, ZNF292, ZNF141, LDB3, PHF14, ZNF322, PDLIM1, NR3C2, and LIMS2) were selected to establish an overall survival (OS) prediction model, and seven ZNFs (ZIC1, ZNF141, ZHX2, ZNF281, ZNHIT2, NR3C2, and LIMS2) were used to develop a progression-free survival (PFS) prediction model. Compared with patients with low-risk in the TCGA training and testing cohorts, as well as the GEO validation cohorts, patients with high-risk had poorer OS and PFS. Using nomograms constructed with the identified ZNFs predicting OS and PFS, we established a clinically useful model. Four distinct molecular subtypes with different prognostic and immune infiltration characteristics were identified. In vitro experiments showed that ZNF141 promoted the proliferation and viability of STS cells. In conclusion, ZNF-related models are useful as prognostic biomarkers, suggesting their potentials as therapeutic targets in STS. These findings will enable us to develop novel strategies treating STS, which will potentially improve outcomes of patients with STS.
| S-EPMC10188339 | biostudies-literature
Project description:BackgroundVarious fusion strategies (feature-level fusion, matrix-level fusion, and image-level fusion) were used to fuse PET and MR images, which might lead to different feature values and classification performance. The purpose of this study was to measure the classification capability of features extracted using various PET/MR fusion methods in a dataset of soft-tissue sarcoma (STS).MethodsThe retrospective dataset included 51 patients with histologically proven STS. All patients had pre-treatment PET and MR images. The image-level fusion was conducted using discrete wavelet transformation (DWT). During the DWT process, the MR weight was set as 0.1, 0.2, 0.3, 0.4, …, 0.9. And the corresponding PET weight was set as 1- (MR weight). The fused PET/MR images was generated using the inverse DWT. The matrix-level fusion was conducted by fusing the feature calculation matrix during the feature extracting process. The feature-level fusion was conducted by concatenating and averaging the features. We measured the predictive performance of features using univariate analysis and multivariable analysis. The univariate analysis included the Mann-Whitney U test and receiver operating characteristic (ROC) analysis. The multivariable analysis was used to develop the signatures by jointing the maximum relevance minimum redundancy method and multivariable logistic regression. The area under the ROC curve (AUC) value was calculated to evaluate the classification performance.ResultsBy using the univariate analysis, the features extracted using image-level fusion method showed the optimal classification performance. For the multivariable analysis, the signatures developed using the image-level fusion-based features showed the best performance. For the T1/PET image-level fusion, the signature developed using the MR weight of 0.1 showed the optimal performance (0.9524(95% confidence interval (CI), 0.8413-0.9999)). For the T2/PET image-level fusion, the signature developed using the MR weight of 0.3 showed the optimal performance (0.9048(95%CI, 0.7356-0.9999)).ConclusionsFor the fusion of PET/MR images in patients with STS, the signatures developed using the image-level fusion-based features showed the optimal classification performance than the signatures developed using the feature-level fusion and matrix-level fusion-based features, as well as the single modality features. The image-level fusion method was more recommended to fuse PET/MR images in future radiomics studies.
| S-EPMC8756708 | biostudies-literature
Project description:PurposeTo develop a rapid 2D MR fingerprinting technique with a submillimeter in-plane resolution using a deep learning-based tissue quantification approach.MethodsA rapid and high-resolution MR fingerprinting technique was developed for brain T1 and T2 quantification. The 2D acquisition was performed using a FISP-based MR fingerprinting sequence and a spiral trajectory with 0.8-mm in-plane resolution. A deep learning-based method was used to replace the standard template matching method for improved tissue characterization. A novel network architecture (i.e., residual channel attention U-Net) was proposed to improve high-resolution details in the estimated tissue maps. Quantitative brain imaging was performed with 5 adults and 2 pediatric subjects, and the performance of the proposed approach was compared with several existing methods in the literature.ResultsIn vivo measurements with both adult and pediatric subjects show that high-quality T1 and T2 mapping with 0.8-mm in-plane resolution can be achieved in 7.5 seconds per slice. The proposed deep learning method outperformed existing algorithms in tissue quantification with improved accuracy. Compared with the standard U-Net, high-resolution details in brain tissues were better preserved by the proposed residual channel attention U-Net. Experiments on pediatric subjects further demonstrated the potential of the proposed technique for fast pediatric neuroimaging. Alongside reduced data acquisition time, a 5-fold acceleration in tissue property mapping was also achieved with the proposed method.ConclusionA rapid and high-resolution MR fingerprinting technique was developed, which enables high-quality T1 and T2 quantification with 0.8-mm in-plane resolution in 7.5 seconds per slice.
| S-EPMC8210473 | biostudies-literature
Project description:The Gleason grading system remains the most powerful prognostic predictor for patients with prostate cancer since the 1960s. Its application requires highly-trained pathologists, is tedious and yet suffers from limited inter-pathologist reproducibility, especially for the intermediate Gleason score 7. Automated annotation procedures constitute a viable solution to remedy these limitations. In this study, we present a deep learning approach for automated Gleason grading of prostate cancer tissue microarrays with Hematoxylin and Eosin (H&E) staining. Our system was trained using detailed Gleason annotations on a discovery cohort of 641 patients and was then evaluated on an independent test cohort of 245 patients annotated by two pathologists. On the test cohort, the inter-annotator agreements between the model and each pathologist, quantified via Cohen's quadratic kappa statistic, were 0.75 and 0.71 respectively, comparable with the inter-pathologist agreement (kappa = 0.71). Furthermore, the model's Gleason score assignments achieved pathology expert-level stratification of patients into prognostically distinct groups, on the basis of disease-specific survival data available for the test cohort. Overall, our study shows promising results regarding the applicability of deep learning-based solutions towards more objective and reproducible prostate cancer grading, especially for cases with heterogeneous Gleason patterns.
| S-EPMC6089889 | biostudies-literature