Automated Prediction of Early Recurrence in Advanced Sinonasal Squamous Cell Carcinoma With Deep Learning and Multi-parametric MRI-based Radiomics Nomogram

Deep Learning Model for the Differential Diagnosis of Nasal Polyps and Inverted Papilloma by CT Images: A Multicenter Study

RATIONALE AND OBJECTIVES

Nasal polyps (NP) and inverted papilloma (IP) are benign tumors within the nasal cavity, each necessitating distinct treatment approaches. Herein, we investigate the utility of a deep learning (DL) model for distinguishing between NP and IP.

MATERIALS AND METHODS

A total of 1791 patients with nasal benign tumors from two hospitals were retrospectively enrolled. Patients were divided into training, internal test, and external test sets. DL models (3D ResNet, 3D Xception, and HRNet) were employed to identify NP from IP using computed tomography images. Model performance was evaluated via receiver operating characteristic curve analysis, accuracy, sensitivity, and specificity. The best-performing model was compared with radiologists' interpretations. The potential enhancement of radiologists' diagnostic performance using the optimal DL model was investigated. Additionally, proteomics analysis in 70 patients was conducted to elucidate the biological underpinnings of the DL model.

RESULTS

The 3D Xception model emerged as the best-performing DL model, achieving the highest area under the receiver operating characteristic curve of 0.999 (95% confidence interval [CI]: 0.950-1.000) in the training set, 0.981 (95% CI: 0.950-1.000) in the internal test set, and 0.933 (95% CI: 0.9099-0.9557) in the external test set. The sensitivity and specificity of the optimal DL model surpassed those of the four radiologists. Furthermore, the DL model improved average radiologist sensitivity from 0.845 to 0.884 and specificity from 0.670 to 0.840. Proteomic analysis revealed an association between the model predictions and epithelial cell differentiation.

CONCLUSION

DL based on CT images holds promise for distinguishing between NP and IP lesions, thereby augmenting clinicians' interpretation capabilities.

Recurrent Sinonasal Squamous Cell Carcinoma: Current Insights and Treatment Advances

Squamous cell carcinoma is the most common malignancy affecting the sinonasal tract. Local recurrence is the main pattern of treatment failure, affecting nearly half of patients treated for primary sinonasal squamous cell carcinoma (SNSCC). Due to disease rarity and heterogeneity of practices, there are limited guidelines for how to diagnose and care for these patients. This paper reviews current evidence regarding etiology, pathophysiology, diagnosis, prognostic factors, and treatment modalities of recurrent SNSCC (rSNSCC). Currently, salvage surgery offers the only durable approach for eligible patients. These resections often require robust reconstructive options due to prior surgery or radiation. Chemoradiation is offered as an adjuvant or palliative approach when surgery is not a feasible option. Emerging options such as immunotherapy and particle therapy remain an area of ongoing investigation.

Value of radiomics and deep learning feature fusion models based on dce-mri in distinguishing sinonasal squamous cell carcinoma from lymphoma

Problem

Sinonasal squamous cell carcinoma (SNSCC) and sinonasal lymphoma (SNL) lack distinct clinical manifestations and traditional imaging characteristics, complicating the accurate differentiation between these tumors and the selection of appropriate treatment strategies. Consequently, there is an urgent need for a method that can precisely distinguish between these tumors preoperatively to formulate suitable treatment plans for patients.

Methods

This study aims to construct and validate ML and DL feature models based on Dynamic Contrast-Enhanced (DCE) imaging and to evaluate the clinical value of a radiomics and deep learning (DL) feature fusion model in differentiating between SNSCC and SNL. This study performed a retrospective analysis on the preoperative axial DCE-T1WI MRI images of 90 patients diagnosed with sinonasal tumors, comprising 50 cases of SNSCC and 40 cases of SNL. Data were randomly divided into a training set and a validation set at a 7:3 ratio, and radiomic features were extracted. Concurrently, deep learning features were derived using the optimally pre-trained DL model and integrated with manually extracted radiomic features. Feature sets were selected through independent samples t-test, Mann-Whitney U-test, Pearson correlation coefficient and LASSO regression. Three conventional machine learning (CML) models and three DL models were established, and all radiomic and DL features were merged to create three pre-fusion machine learning models (DLR). Additionally, a post-fusion model (DLRN) was constructed by combining radiomic scores and DL scores. Quantitative metrics such as area under the curve (AUC), sensitivity, and accuracy were employed to identify the optimal feature set and classifier. Furthermore, a deep learning-radiomics nomogram (DLRN) was developed as a clinical decision-support tool.

Results

The feature fusion model of radiomics and DL has higher accuracy in distinguishing SNSCC from SNL than CML or DL alone. The ExtraTrees model based on DLR fusion features of DCE-T1WI had an AUC value of 0.995 in the training set and 0.939 in the validation set.The DLRN model based on the fusion of predictive scores had an AUC value of 0.995 in the training set and 0.911 in the validation set.The DLRN model based on the fusion of predictive scores had an AUC value of 0.995 in the training set and 0.911 in the validation set.

Conclusion

This study, by constructing a feature integration model combining radiomics and deep learning (DL), has demonstrated strong predictive capabilities in the preoperative non-invasive diagnosis of SNSCC and SNL, offering valuable information for tailoring personalized treatment plans for patients.

Computer Vision-Radiomics & Pathognomics

The role of computer vision in extracting radiographic (radiomics) and histopathologic (pathognomics) features is an extension of molecular biomarkers that have been foundational to our understanding across the spectrum of head and neck disorders. Especially within head and neck cancers, machine learning and deep learning applications have yielded advances in the characterization of tumor features, nodal features, and various outcomes. This review aims to overview the landscape of radiomic and pathognomic applications, informing future work to address gaps. Novel methodologies will be needed to potentially engineer ways of integrating multidimensional data inputs to examine disease features to guide prognosis comprehensively and ultimately clinical management.

Effect of fully automatic classification model from different tube voltage images on bone density screening: A self-controlled study

PURPOSE

To develop two bone status prediction models combining deep learning and radiomics based on standard-dose chest computed tomography (SDCT) and low-dose chest computed tomography (LDCT), and to evaluate the effect of tube voltage on reproducibility of radiomics features and predictive efficacy of these models.

METHODS

A total of 1508 patients were enrolled in this retrospective study. LDCT was conducted using 80 kVp, tube current ranging from 100 to 475 mA. On the other hand, SDCT was performed using 120 kVp, tube current ranging from 100 to 520 mA. We developed an automatic thoracic vertebral cancellous bone (TVCB) segmentation model. Subsequently, 1184 features were extracted and two classifiers were developed based on LDCT and SDCT images. Based on the diagnostic results of quantitative computed tomography examination, the first-level classifier was initially developed to distinguish normal or abnormal BMD (including osteoporosis and osteopenia), while the second-level classifier was employed to identify osteoporosis or osteopenia. The Dice coefficient was used to evaluate the performance of the automated segmentation model. The Concordance Correlation Coefficients (CCC) of radiomics features were calculated between LDCT and SDCT, and the performance of these models was evaluated.

RESULTS

Our automated segmentation model achieved a Dice coefficient of 0.98 ± 0.01 and 0.97 ± 0.02 in LDCT and SDCT, respectively. Alterations in tube voltage decreased the reproducibility of the extracted radiomic features, with 85.05 % of the radiomic features exhibiting low reproducibility (CCC < 0.75). The area under the curve (AUC) using LDCT-based and SDCT-based models was 0.97 ± 0.01 and 0.94 ± 0.02, respectively. Nonetheless, cross-validation with independent test sets of different tube voltage scans suggests that variations in tube voltage can impair the diagnostic efficacy of the model. Consequently, radiomics models are not universally applicable to images of varying tube voltages. In clinical settings, ensuring consistency between the tube voltage of the image used for model development and that of the acquired patient image is critical.

CONCLUSIONS

Automatic bone status prediction models, utilizing either LDCT or SDCT images, enable accurate assessment of bone status. Tube voltage impacts reproducibility of features and predictive efficacy of models. It is necessary to account for tube voltage variation during the image acquisition.

MRI- and DWI-Based Radiomics Features for Preoperatively Predicting Meningioma Sinus Invasion

The aim of this study was to use multimodal imaging (contrast-enhanced T1-weighted (T1C), T2-weighted (T2), and diffusion-weighted imaging (DWI)) to develop a radiomics model for preoperatively predicting venous sinus invasion in meningiomas. This prediction would assist in selecting the appropriate surgical approach and forecasting the prognosis of meningiomas. A retrospective analysis was conducted on 331 participants who had been pathologically diagnosed with meningiomas. For each participant, 3948 radiomics features were acquired from the T1C, T2, and DWI images. Minimum redundancy maximum correlation, rank sum test, and multi-factor recursive elimination were used to extract the most significant features of different models. Then, multivariate logistic regression was used to build classification models to predict meningioma venous sinus invasion. The diagnostic capabilities were assessed using receiver operating characteristic (ROC) analysis. In addition, a nomogram was constructed by incorporating clinical and radiological characteristics and a radiomics signature. To assess the clinical usefulness of the nomogram, a decision curve analysis (DCA) was performed. Tumor shape, boundary, and enhancement features were independent predictors of meningioma venous sinus invasion (p = 0.013, p = 0.013, p = 0.005, respectively). Eleven (T2:1, T1C:4, DWI:6) of the 3948 radiomics features were screened for strong association with meningioma sinus invasion. The areas under the ROC curves for the training and external test sets were 0.946 and 0.874, respectively. The clinicoradiomic model showed excellent predictive performance for invasive meningioma, which may help to guide surgical approaches and predict prognosis.

Machine Learning-Based MRI Radiogenomics for Evaluation of Response to Induction Chemotherapy in Head and Neck Squamous Cell Carcinoma

RATIONALE AND OBJECTIVES

To develop and validate a radiogenomics model integrating clinical data, radiomics-based machine learning (RBML) classifiers, and transcriptomics data for predicting the response to induction chemotherapy (IC) in patients with head and neck squamous cell carcinoma (HNSCC).

MATERIALS AND METHODS

Radiomics features derived from T2-weighted, pre- and post-contrast-enhanced T1-weighted MRI sequences, clinical data, and RNA sequencing data of 150 patients with HNSCC were included in the study. Analysis of variance or recursive feature elimination was used to reduce radiomics features. Three RBML classifiers were developed to distinguish non-responders from responders. Weighted correlation network analysis (WGCNA) was performed to identify the correlation between clinical data or radiomics features and molecular features; subsequently, protein interaction and functional enrichment analyses were performed. The predictive performance of the radiogenomics model integrating significant clinical variables, RBML classifiers, and molecular features was evaluated using receiver operating characteristic curve analysis.

RESULTS

Five radiomics features and two conventional MRI findings significantly stratified HNSCC patients into responders and non-responders. On WGCNA analysis, 809 genes showed a significant correlation with two radiomics features. Functional enrichment analysis suggested that our proposed radiomics features could reflect the T cell-mediated immune response and immune infiltration of HNSCC. The radiogenomics model showed the highest area under the curve (0.88[95%CI 0.75-0.96]) for predicting IC response, which was better than MRI findings(p = 0.0407) or molecular features(p = 0.004) alone, but showed no significant difference with that of RBML model (p = 0.2254) in test cohort.

CONCLUSION

Merging imaging phenotypes with transcriptomic data improved the prediction of IC response in HNSCC.

Automatic Osteoporosis Screening System Using Radiomics and Deep Learning from Low-Dose Chest CT Images

OBJECTIVE

Develop two fully automatic osteoporosis screening systems using deep learning (DL) and radiomics (Rad) techniques based on low-dose chest CT (LDCT) images and evaluate their diagnostic effectiveness.

METHODS

In total, 434 patients who underwent LDCT and bone mineral density (BMD) examination were retrospectively enrolled and divided into the development set (n = 333) and temporal validation set (n = 101). An automatic thoracic vertebra cancellous bone (TVCB) segmentation model was developed. The Dice similarity coefficient (DSC) was used to evaluate the segmentation performance. Furthermore, the three-class Rad and DL models were developed to distinguish osteoporosis, osteopenia, and normal bone mass. The diagnostic performance of these models was evaluated using the receiver operating characteristic (ROC) curve and decision curve analysis (DCA).

RESULTS

The automatic segmentation model achieved excellent segmentation performance, with a mean DSC of 0.96 ± 0.02 in the temporal validation set. The Rad model was used to identify osteoporosis, osteopenia, and normal BMD in the temporal validation set, with respective area under the receiver operating characteristic curve (AUC) values of 0.943, 0.801, and 0.932. The DL model achieved higher AUC values of 0.983, 0.906, and 0.969 for the same categories in the same validation set. The Delong test affirmed that both models performed similarly in BMD assessment. However, the accuracy of the DL model is 81.2%, which is better than the 73.3% accuracy of the Rad model in the temporal validation set. Additionally, DCA indicated that the DL model provided a greater net benefit compared to the Rad model across the majority of the reasonable threshold probabilities Conclusions: The automated segmentation framework we developed can accurately segment cancellous bone on low-dose chest CT images. These predictive models, which are based on deep learning and radiomics, provided comparable diagnostic performance in automatic BMD assessment. Nevertheless, it is important to highlight that the DL model demonstrates higher accuracy and precision than the Rad model.