Predictive Imaging Features Research Articles

Differentiation Between Invasive Adenocarcinoma and Focal Interstitial Fibrosis among Persistent Pulmonary Part-solid Nodules: With Emphasis on the CT Morphologic Analysis.

Focal interstitial fibrosis (FIF) manifesting as a persistent part-solid nodule (PSN) has been mistakenly treated surgically due to similar imaging features to invasive adenocarcinoma (ADC). The purpose of this study was to observe predictive imaging features correlated with FIF through CT morphologic analysis. From January 2009 to December 2020, 44 patients with surgically proven FIF in a single institution were enrolled and compared with 88 ADC patients through propensity score matching. Patient characteristics and CT morphologic analysis of persistent PSNs were used to identify predictive imaging features of FIF. Receiver operating characteristic (ROC) curve analysis was used to quantify the performance of imaging features. A total of 132 patients with 132 PSNs (44 FIF, 88 ADC; mean age, 67.7±7.58; 75 females) were involved in our analysis. Multivariable analysis demonstrated that preserved peritumoral vascular margin (preserved vascular margin), preserved secondary pulmonary lobule margin (preserved lobular margin), and lower coronal to axial ratio (C/A ratio; cutoff: 1.005) were significant independent predictors of FIF ( P< 0.05). ROC curve analysis to evaluate the predictive value of the logistic model based on the imaging features of FIF, and the AUC value was 0.881. CT imaging features of preserved vascular margin, preserved lobular margin, and lower C/A ratio (cutoff, <1.005) might be helpful imaging features in discriminating FIF over ADC among persistent PSN in clinical practice.

Lesion attention guided neural network for contrast-enhanced mammography-based biomarker status prediction in breast cancer

Background and objectiveAccurate identification of molecular biomarker statuses is crucial in cancer diagnosis, treatment, and prognosis. Studies have demonstrated that medical images could be utilized for non-invasive prediction of biomarker statues. The biomarker status-associated features extracted from medical images are essential in developing medical image-based non-invasive prediction models. Contrast-enhanced mammography (CEM) is a promising imaging technique for breast cancer diagnosis. This study aims to develop a neural network-based method to extract biomarker-related image features from CEM images and evaluate the potential of CEM in non-invasive biomarker status prediction. MethodsAn end-to-end learning convolutional neural network with the whole breast images as inputs was proposed to extract CEM features for biomarker status prediction in breast cancer. The network focused on lesion regions and flexibly extracted image features from lesion and peri‑tumor regions by employing supervised learning with a smooth L1-based consistency constraint. An image-level weakly supervised segmentation network based on Vision Transformer with cross attention to contrast images of breasts with lesions against the contralateral breast images was developed for automatic lesion segmentation. Finally, prediction models were developed following further selection of significant features and the implementation of random forest-based classification. Results were reported using the area under the curve (AUC), accuracy, sensitivity, and specificity. ResultsA dataset from 1203 breast cancer patients was utilized to develop and evaluate the proposed method. Compared to the method without lesion attention and with only lesion regions as inputs, the proposed method performed better at biomarker status prediction. Specifically, it achieved an AUC of 0.71 (95 % confidence interval [CI]: 0.65, 0.77) for Ki-67 and 0.73 (95 % CI: 0.65, 0.80) for human epidermal growth factor receptor 2 (HER2). ConclusionsA lesion attention-guided neural network was proposed in this work to extract CEM image features for biomarker status prediction in breast cancer. The promising results demonstrated the potential of CEM in non-invasively predicting the biomarker statuses in breast cancer.

Brain magnetic resonance imaging outperforms clinical severity ratings in the prediction of treatment outcomes in major depressive disorder

IntroductionMajor depressive disorder (MDD) is a prevalent and disabling condition. Approximately 30-50% of patients do not respond to first-line medication or psychotherapy. Therefore, several studies have investigated the predictive potential of pretreatment severity rating or neuroimaging features to guide clinical approaches that can speed optimal treatment selection.ObjectivesTo evaluate the performance of 1) severity ratings (scores of Hamilton Depression/Anxiety Scale, illness duration, and sleep quality, etc.) and demographic characteristic and 2) brain magnetic resonance imaging (MRI) features in predicting treatment outcomes for MDD. Second, to assess performance variations among varied modalities and interventions in MRI studies.MethodsWe searched studies in PubMed, Embase, Web of Science, and Science Direct databases before March 22, 2023. We extracted a confusion matrix for prediction in each study. Separate meta-analyses were performed for clinical and MRI studies. The logarithm of diagnostic odds ratio [log(DOR)], sensitivity, and specificity were conducted using Reitsma’s random effect model. The area under curve (AUC) of summary receiver operating characteristic (SROC) curve was calculated.Subgroup analyses were conducted in MRI studies based on modalities: resting-state functional MRI (rsfMRI), task-based fMRI (tbfMRI), and structural MRI (sMRI), and interventions: antidepressant (including selective serotonin reuptake inhibitors [SSRI]) and electroconvulsive therapy (ECT). Meta-regression was conducted 1) between clinical and MRI studies and 2) among modality or intervention subgroups in MRI studies.ResultsWe included ten studies used clinical features covering 6494 patients, yielded a log(DOR) of 1.42, AUC of 0.71, sensitivity of 0.61, and specificity of 0.74. In terms of MRI, 44 studies with 2623 patients were included, revealing an overall log(DOR) of 2.53. The AUC, sensitivity, and specificity were 0.89, 0.78, and 0.75.Studies using MRI features had a higher sensitivity (0.89 vs. 0.61) in predicting treatment outcomes than clinical features (P &lt; 0.001). RsfMRI had higher specificity (0.79 vs. 0.69) than tbfMRI subgroup (P = 0.01). No significant differences were found between sMRI and other modalities, nor between antidepressants (SSRIs and others) and ECT. Antidepressant studies primarily identified predictive imaging features in limbic and default mode networks, while ECT mainly focused on limbic network.ConclusionsOur findings suggest a robust promise for pretreatment brain MRI features in predicting treatment outcomes in MDD, offering higher accuracy than clinical studies. While tasks in tbfMRI studies differed, those studies overall had less predictive utility than rsfMRI data. For MRI studies, overlapping but distinct network level measures predicted outcomes for antidepressants and ECT.Disclosure of InterestNone Declared

MPS-NeRF: Generalizable 3D Human Rendering From Multiview Images.

There has been rapid progress recently on 3D human rendering, including novel view synthesis and pose animation, based on the advances of neural radiance fields (NeRF). However, most existing methods focus on person-specific training and their training typically requires multi-view videos. This paper deals with a new challenging task - rendering novel views and novel poses for a person unseen in training, using only multiview still images as input without videos. For this task, we propose a simple yet surprisingly effective method to train a generalizable NeRF with multiview images as conditional input. The key ingredient is a dedicated representation combining a canonical NeRF and a volume deformation scheme. Using a canonical space enables our method to learn shared properties of human and easily generalize to different people. Volume deformation is used to connect the canonical space with input and target images and query image features for radiance and density prediction. We leverage the parametric 3D human model fitted on the input images to derive the deformation, which works quite well in practice when combined with our canonical NeRF. The experiments on both real and synthetic data with the novel view synthesis and pose animation tasks collectively demonstrate the efficacy of our method.

Integrating spectral and image information for prediction of cottonseed vitality.

Cotton plays a significant role in people's lives, and cottonseeds serve as a vital assurance for successful cotton cultivation and production. Premium-quality cottonseeds can significantly enhance the germination rate of cottonseeds, resulting in increased cotton yields. The vitality of cottonseeds is a crucial metric that reflects the quality of the seeds. However, currently, the industry lacks a non-destructive method to directly assess cottonseed vitality without compromising the integrity of the seeds. To address this challenge, this study employed a hyperspectral imaging acquisition system to gather hyperspectral data on cottonseeds. This system enables the simultaneous collection of hyperspectral data from 25 cottonseeds. This study extracted spectral and image information from the hyperspectral data of cottonseeds to predict their vitality. SG, SNV, and MSC methods were utilized to preprocess the spectral data of cottonseeds. Following this preprocessing step, feature wavelength points of the cottonseeds were extracted using SPA and CARS algorithms. Subsequently, GLCM was employed to extract texture features from images corresponding to these feature wavelength points, including attributes such as Contrast, Correlation, Energy, and Entropy. Finally, the vitality of cottonseeds was predicted using PLSR, SVR, and a self-built 1D-CNN model. For spectral data analysis, the 1D-CNN model constructed after MSC+CARS preprocessing demonstrated the highest performance, achieving a test set correlation coefficient of 0.9214 and an RMSE of 0.7017. For image data analysis, the 1D-CNN model constructed after SG+CARS preprocessing outperformed the others, yielding a test set correlation coefficient of 0.8032 and an RMSE of 0.9683. In the case of fused spectral and image data, the 1D-CNN model built after SG+SPA preprocessing displayed the best performance, attaining a test set correlation coefficient of 0.9427 and an RMSE of 0.6872. These findings highlight the effectiveness of the 1D-CNN model and the fusion of spectral and image features for cottonseed vitality prediction. This research contributes significantly to the development of automated detection devices for assessing cottonseed vitality.

Unmasking air quality: A novel image-based approach to align public perception with pollution levels

In the quest to reconcile public perception of air pollution with scientific measurements, our study introduced a pioneering method involving a gradient boost-regression tree model integrating PM2.5 concentration, visibility, and image-based data. Traditional stationary monitoring often falls short of accurately capturing public air quality perceptions, prompting the need for alternative strategies. Leveraging an extensive dataset of over 20,000 public visibility perception evaluations and over 8,000 stationary images, our models effectively quantify diverse air quality perceptions. The predictive prowess of our models was validated by strong performance metrics for perceived visibility (R = 0.98, RMSE = 0.19), all-day PM2.5 concentrations (R: 0.77–0.78, RMSE: 8.31–9.40), and Central Weather Bureau visibility records (R = 0.82, RMSE = 9.00). Interestingly, image contrast and light intensity hold greater importance than scenery clarity in the visibility perception model. However, clarity is prioritized in PM2.5 and Central Weather Bureau models. Our research also unveiled spatial limitations in stationary monitoring and outlined the variations in predictive image features between near and far stations. Crucially, all models benefit from the characterization of atmospheric light sources through defogging techniques. The image-based insights highlight the disparity between public perception of air pollution and current policy implementation. In other words, policymakers should shift from solely emphasizing the reduction of PM2.5 levels to also incorporating the public's perception of visibility into their strategies. Our findings have broad implications for air quality evaluation, image mining in specific areas, and formulating air quality management strategies that account for public perception.

Predictive clinical and imaging features of high-grade dysplasia in mucinous cystic neoplasms of the pancreas

Predictive clinical and imaging features of high-grade dysplasia in mucinous cystic neoplasms of the pancreas

Comparison of computed tomography image features extracted by radiomics, self-supervised learning and end-to-end deep learning for outcome prediction of oropharyngeal cancer

Background and purposeTo compare the prediction performance of image features of computed tomography (CT) images extracted by radiomics, self-supervised learning and end-to-end deep learning for local control (LC), regional control (RC), locoregional control (LRC), distant metastasis-free survival (DMFS), tumor-specific survival (TSS), overall survival (OS) and disease-free survival (DFS) of oropharyngeal squamous cell carcinoma (OPSCC) patients after (chemo)radiotherapy. Methods and materialsThe OPC-Radiomics dataset was used for model development and independent internal testing and the UMCG-OPC set for external testing. Image features were extracted from the Gross Tumor Volume contours of the primary tumor (GTVt) regions in CT scans when using radiomics or a self-supervised learning-based method (autoencoder). Clinical and combined (radiomics, autoencoder or end-to-end) models were built using multivariable Cox proportional-hazard analysis with clinical features only and both clinical and image features for LC, RC, LRC, DMFS, TSS, OS and DFS prediction, respectively. ResultsIn the internal test set, combined autoencoder models performed better than clinical models and combined radiomics models for LC, RC, LRC, DMFS, TSS and DFS prediction (largest improvements in C-index: 0.91 vs. 0.76 in RC and 0.74 vs. 0.60 in DMFS). In the external test set, combined radiomics models performed better than clinical and combined autoencoder models for all endpoints (largest improvements in LC, 0.82 vs. 0.71). Furthermore, combined models performed better in risk stratification than clinical models and showed good calibration for most endpoints. ConclusionsImage features extracted using self-supervised learning showed best internal prediction performance while radiomics features have better external generalizability.

Survival prediction for stage I-IIIA non-small cell lung cancer using deep learning

Background and purposeThe aim of this study was to develop and evaluate a prediction model for 2-year overall survival (OS) in stage I-IIIA non-small cell lung cancer (NSCLC) patients who received definitive radiotherapy by considering clinical variables and image features from pre-treatment CT-scans. Materials and methodsNSCLC patients who received stereotactic radiotherapy were prospectively collected at the UMCG and split into a training and a hold out test set including 189 and 81 patients, respectively. External validation was performed on 228 NSCLC patients who were treated with radiation or concurrent chemoradiation at the Maastro clinic (Lung1 dataset). A hybrid model that integrated both image and clinical features was implemented using deep learning. Image features were learned from cubic patches containing lung tumours extracted from pre-treatment CT scans. Relevant clinical variables were selected by univariable and multivariable analyses. ResultsMultivariable analysis showed that age and clinical stage were significant prognostic clinical factors for 2-year OS. Using these two clinical variables in combination with image features from pre-treatment CT scans, the hybrid model achieved a median AUC of 0.76 [95 % CI: 0.65–0.86] and 0.64 [95 % CI: 0.58–0.70] on the complete UMCG and Maastro test sets, respectively. The Kaplan-Meier survival curves showed significant separation between low and high mortality risk groups on these two test sets (log-rank test: p-value < 0.001, p-value = 0.012, respectively) ConclusionWe demonstrated that a hybrid model could achieve reasonable performance by utilizing both clinical and image features for 2-year OS prediction. Such a model has the potential to identify patients with high mortality risk and guide clinical decision making.

Identification of fibrosis in hypertrophic cardiomyopathy: a radiomic study on cardiac magnetic resonance cine imaging

ObjectivesHypertrophic cardiomyopathy (HCM) often requires repeated enhanced cardiac magnetic resonance (CMR) imaging to detect fibrosis. We aimed to develop a practical model based on cine imaging to help identify patients with high risk of fibrosis and screen out patients without fibrosis to avoid unnecessary injection of contrast.MethodsA total of 273 patients with HCM were divided into training and test sets at a ratio of 7:3. Logistic regression analysis was used to find predictive image features to construct CMR model. Radiomic features were derived from the maximal wall thickness (MWT) slice and entire left ventricular (LV) myocardium. Extreme gradient boosting was used to build radiomic models. Integrated models were established by fusing image features and radiomic models. The model performance was validated in the test set and assessed by ROC and calibration curve and decision curve analysis (DCA).ResultsWe established five prediction models, including CMR, R1 (based on the MWT slice), R2 (based on the entire LV myocardium), and two integrated models (ICMR+R1 and ICMR+R2). In the test set, ICMR+R2 model had an excellent AUC value (0.898), diagnostic accuracy (89.02%), sensitivity (92.54%), and F1 score (93.23%) in identifying patients with positive late gadolinium enhancement. The calibration plots and DCA indicated that ICMR+R2 model was well-calibrated and presented a better net benefit than other models.ConclusionsA predictive model that fused image and radiomic features from the entire LV myocardium had good diagnostic performance, robustness, and clinical utility.Key Points• Hypertrophic cardiomyopathy is prone to fibrosis, requiring patients to undergo repeated enhanced cardiac magnetic resonance imaging to detect fibrosis over their lifetime follow-up.• A predictive model based on the entire left ventricular myocardium outperformed a model based on a slice of the maximal wall thickness.• A predictive model that fused image and radiomic features from the entire left ventricular myocardium had excellent diagnostic performance, robustness, and clinical utility.

Differentiating central nervous system infection from disease infiltration in hematological malignancy

Hematological malignancies place individuals at risk of CNS involvement from their hematological disease and opportunistic intracranial infection secondary to disease-/treatment-associated immunosuppression. Differentiating CNS infection from hematological disease infiltration in these patients is valuable but often challenging. We sought to determine if statistical models might aid discrimination between these processes. Neuroradiology, clinical and laboratory data for patients with hematological malignancy at our institution between 2007 and 2017 were retrieved. MRI were deep-phenotyped across anatomical distribution, presence of pathological enhancement, diffusion restriction and hemorrhage and statistically modelled with Bayesian-directed probability networks and multivariate logistic regression. 109 patients were studied. Irrespective of a diagnosis of CNS infection or hematological disease, the commonest anatomical distributions of abnormality were multifocal-parenchymal (34.9%), focal-parenchymal (29.4%) and leptomeningeal (11.9%). Pathological enhancement was the most frequently observed abnormality (46.8%), followed by hemorrhage (22.9%) and restricted diffusion (19.3%). Logistic regression could differentiate CNS infection from hematological disease infiltration with an AUC of 0.85 where, with OR > 1 favoring CNS infection and < 1 favoring CNS hematological disease, significantly predictive imaging features were hemorrhage (OR 24.61, p = 0.02), pathological enhancement (OR 0.17, p = 0.04) and an extra-axial location (OR 0.06, p = 0.05). In conclusion, CNS infection and hematological disease are heterogeneous entities with overlapping radiological appearances but a multivariate interaction of MR imaging features may assist in distinguishing them.

Neoadjuvant Chemoradiotherapy for Oral Cavity Cancer: Predictive Factors for Response and Interim Analysis of the Prospective INVERT-Trial.

BackgroundTo study neoadjuvant chemoradiotherapy (nCRT) and potential predictive factors for response in locally advanced oral cavity cancer (LA-OCC).MethodsThe INVERT trial is an ongoing single-center, prospective phase 2, proof-of-principle trial. Operable patients with stage III-IVA squamous cell carcinomas of the oral cavity were eligible and received nCRT consisting of 60 Gy with concomitant cisplatin and 5-fluorouracil. Surgery was scheduled 6-8 weeks after completion of nCRT. Explorative, multiplex immunohistochemistry (IHC) was performed on pretreatment tumor specimen, and diffusion-weighted magnetic resonance imaging (DW-MRI) was conducted prior to, during nCRT (day 15), and before surgery to identify potential predictive biomarkers and imaging features. Primary endpoint was the pathological complete response (pCR) rate.ResultsSeventeen patients with stage IVA OCC were included in this interim analysis. All patients completed nCRT. One patient died from pneumonia 10 weeks after nCRT before surgery. Complete tumor resection (R0) was achieved in 16/17 patients, of whom 7 (41%, 95% CI: 18-67%) showed pCR. According to the Clavien-Dindo classification, grade 3a and 3b complications were found in 4 (25%) and 5 (31%) patients, respectively; grade 4-5 complications did not occur. Increased changes in the apparent diffusion coefficient signal intensities between MRI at day 15 of nCRT and before surgery were associated with better response (p=0.022). Higher abundances of programmed cell death protein 1 (PD1) positive cytotoxic T-cells (p=0.012), PD1+ macrophages (p=0.046), and cancer-associated fibroblasts (CAFs, p=0.036) were associated with incomplete response to nCRT.ConclusionnCRT for LA-OCC followed by radical surgery is feasible and shows high response rates. Larger patient cohorts from randomized trials are needed to further investigate nCRT and predictive biomarkers such as changes in DW-MRI signal intensities, tumor infiltrating immune cells, and CAFs.

Imaging and Clinical Parameters for Distinction between Infected and Non-Infected Fluid Collections in CT: Prospective Study Using Extended Microbiological Approach.

The aim of this investigation was to evaluate predictive CT imaging features and clinical parameters to distinguish infected from sterile fluid collections. Detection of infectious agents by advanced microbiological analysis was used as the reference standard. From April 2018 to October 2019, all patients undergoing CT-guided drainages were prospectively enrolled, if drainage material volume was at least 5 mL. Univariate analysis revealed attenuation (p = 0.001), entrapped gas (p < 0.001), fat stranding (p < 0.001), wall thickness (p < 0.001) and enhancement (p < 0.001) as imaging biomarkers and procalcitonin (p = 0.003) as clinical predictive parameters for infected fluid collections. On multivariate analysis, attenuation > 10 HU (p = 0.038), presence of entrapped gas (p = 0.027) and wall enhancement (p = 0.028) were independent parameters for distinguishing between infected and non-infected fluids. Gas entrapment had high specificity (93%) but low sensitivity (48%), while wall enhancement had high sensitivity (91%) but low specificity (50%). CT attenuation > 10 HU showed intermediate sensitivity (74%) and specificity (70%). Evaluation of the published proposed scoring systems did not improve diagnostic accuracy over independent predictors in our study. In conclusion, this prospective study confirmed that CT attenuation > 10 HU, entrapped gas and wall enhancement are the key imaging features to distinguish infected from sterile fluid collections on CT.

Predictive CT features for the diagnosis of primary pulmonary mucoepidermoid carcinoma: comparison with squamous cell carcinomas and adenocarcinomas

BackgroundTo determine the predictive CT imaging features for diagnosis in patients with primary pulmonary mucoepidermoid carcinomas (PMECs).Materials and methodsCT imaging features of 37 patients with primary PMECs, 76 with squamous cell carcinomas (SCCs) and 78 with adenocarcinomas were retrospectively reviewed. The difference of CT features among the PMECs, SCCs and adenocarcinomas was analyzed using univariate analysis, followed by multinomial logistic regression and receiver operating characteristic (ROC) curve analysis.ResultsCT imaging features including tumor size, location, margin, shape, necrosis and degree of enhancement were significant different among the PMECs, SCCs and adenocarcinomas, as determined by univariate analysis (P < 0.05). Only lesion location, shape, margin and degree of enhancement remained independent factors in multinomial logistic regression analysis. ROC curve analysis showed that the area under curve of the obtained multinomial logistic regression model was 0.805 (95%CI: 0.704–0.906).ConclusionThe prediction model derived from location, margin, shape and degree of enhancement can be used for preoperative diagnosis of PMECs.

Mediastinal Lymph Node Metastasis from Head and Neck Cancer: Predictive Factors and Imaging Features.

PurposeTo assess the predictive factors and describe the imaging features of mediastinal lymph node (MLN) metastases in patients with head and neck cancer.Materials and MethodsWe compared the clinical features and disease characteristics (sex, age, site of primary tumor, histologic type, history of prior treatments, TNM stages, and metastasis in cervical LNs) of patients with head and neck cancers between the MLN metastasis and no MLN metastasis groups. We also evaluated the chest CT (distribution and maximum dimension of the largest LN) and PET/CT (maximum standardized uptake value) features of MLN metastases based on the MLN classification.ResultsOf the 470 patients with head and neck cancer, 55 (11.7%) had MLN metastasis, involving 150 mediastinal stations. Hypopharynx cancer, recurrent tumor, T4 stage, N2/N3 stages, and M1 stage were found to be significant predicting factors for MLN metastasis. The most common location of MLN metastasis was ipsilateral station 2 (upper paratracheal LNs, 36.4%), followed by ipsilateral station 11 (interlobar LNs, 27.3%) and ipsilateral station 10 (hilar LNs, 25.5%).ConclusionMetastasis to MLNs should be considered in patients with head and neck cancer, especially in cases that are associated with a hypopharyngeal cancer, recurrent tumor, and high TNM stages.

Laplacian Pyramid Neural Network for Dense Continuous-Value Regression for Complex Scenes.

Many computer vision tasks, such as monocular depth estimation and height estimation from a satellite orthophoto, have a common underlying goal, which is regression of dense continuous values for the pixels given a single image. We define them as dense continuous-value regression (DCR) tasks. Recent approaches based on deep convolutional neural networks significantly improve the performance of DCR tasks, particularly on pixelwise regression accuracy. However, it still remains challenging to simultaneously preserve the global structure and fine object details in complex scenes. In this article, we take advantage of the efficiency of Laplacian pyramid on representing multiscale contents to reconstruct high-quality signals for complex scenes. We design a Laplacian pyramid neural network (LAPNet), which consists of a Laplacian pyramid decoder (LPD) for signal reconstruction and an adaptive dense feature fusion (ADFF) module to fuse features from the input image. More specifically, we build an LPD to effectively express both global and local scene structures. In our LPD, the upper and lower levels, respectively, represent scene layouts and shape details. We introduce a residual refinement module to progressively complement high-frequency details for signal prediction at each level. To recover the signals at each individual level in the pyramid, an ADFF module is proposed to adaptively fuse multiscale image features for accurate prediction. We conduct comprehensive experiments to evaluate a number of variants of our model on three important DCR tasks, i.e., monocular depth estimation, single-image height estimation, and density map estimation for crowd counting. Experiments demonstrate that our method achieves new state-of-the-art performance in both qualitative and quantitative evaluation on the NYU-D V2 and KITTI for monocular depth estimation, the challenging Urban Semantic 3D (US3D) for satellite height estimation, and four challenging benchmarks for crowd counting. These results demonstrate that the proposed LAPNet is a universal and effective architecture for DCR problems.

Pathomic Fusion: An Integrated Framework for Fusing Histopathology and Genomic Features for Cancer Diagnosis and Prognosis.

Cancer diagnosis, prognosis, mymargin and therapeutic response predictions are based on morphological information from histology slides and molecular profiles from genomic data. However, most deep learning-based objective outcome prediction and grading paradigms are based on histology or genomics alone and do not make use of the complementary information in an intuitive manner. In this work, we propose Pathomic Fusion, an interpretable strategy for end-to-end multimodal fusion of histology image and genomic (mutations, CNV, RNA-Seq) features for survival outcome prediction. Our approach models pairwise feature interactions across modalities by taking the Kronecker product of unimodal feature representations, and controls the expressiveness of each representation via a gating-based attention mechanism. Following supervised learning, we are able to interpret and saliently localize features across each modality, and understand how feature importance shifts when conditioning on multimodal input. We validate our approach using glioma and clear cell renal cell carcinoma datasets from the Cancer Genome Atlas (TCGA), which contains paired whole-slide image, genotype, and transcriptome data with ground truth survival and histologic grade labels. In a 15-fold cross-validation, our results demonstrate that the proposed multimodal fusion paradigm improves prognostic determinations from ground truth grading and molecular subtyping, as well as unimodal deep networks trained on histology and genomic data alone. The proposed method establishes insight and theory on how to train deep networks on multimodal biomedical data in an intuitive manner, which will be useful for other problems in medicine that seek to combine heterogeneous data streams for understanding diseases and predicting response and resistance to treatment. Code and trained models are made available at: https://github.com/mahmoodlab/PathomicFusion.

Fusion of FDG-PET Image and Clinical Features for Prediction of Lung Metastasis in Soft Tissue Sarcomas.

Extracting massive features from images to quantify tumors provides a new insight to solve the problem that tumor heterogeneity is difficult to assess quantitatively. However, quantification of tumors by single-mode methods often has defects such as difficulty in features extraction and high computational complexity. The multimodal approach has shown effective application prospects in solving these problems. In this paper, we propose a feature fusion method based on positron emission tomography (PET) images and clinical information, which is used to obtain features for lung metastasis prediction of soft tissue sarcomas (STSs). Random forest method was adopted to select effective features by eliminating irrelevant or redundant features, and then they were used for the prediction of the lung metastasis combined with back propagation (BP) neural network. The results show that the prediction ability of the proposed model using fusion features is better than that of the model using an image or clinical feature alone. Furthermore, a good performance can be obtained using 3 standard uptake value (SUV) features of PET image and 7 clinical features, and its average accuracy, sensitivity, and specificity on all the sets can reach 92%, 91%, and 92%, respectively. Therefore, the fusing features have the potential to predict lung metastasis for STSs.

Predictive factors and radiological findings of adrenohepatic adhesion during laparoscopic adrenalectomy.

PurposeThis retrospective study aimed to identify predictive factors and imaging features of adrenohepatic adhesion found during laparoscopic right adrenalectomy.Materials and MethodsAltogether, 77 patients underwent laparoscopic right adrenalectomy between January 2005 and December 2018. Adrenohepatic adhesion was defined as strict adhesion that required either partial adrenalectomy with coagulation of residual tissue or partial hepatectomy to accomplish complete resection. We assessed their surgical video records to determine if adrenohepatic adhesion was present. Age, sex, body mass index, tumor size, tumor diagnosis and radiological findings (attachment between the liver and the adrenal gland, diameters of the right and left adrenal veins and its ratio) were evaluated as preoperative variables.ResultsAdrenohepatic adhesion was present in 11 of the 77 patients (14.3%). Age, sex, and body mass index were not statistically significant factors. Tumor size was significantly small in adhesion group (14.2 mm vs. 25.9 mm, p=0.02). Attachment to the liver and adrenal gland was frequently seen regardless of the adhesion. The mean right/left adrenal veins diameters ratio was significantly lower in the adhesion group (0.8 vs. 1.1, p=0.01). Multivariate logistic regression analysis demonstrated the right/left adrenal veins diameters ratio was the only significant predictor of adhesion. The sensitivity, specificity, negative predictive value and positive predictive value were 0.82, 0.76, 0.43, and 0.95 respectively when the optimal cutoff value for the ratio was 0.9 (area under the curve, 0.75; 95% confidence interval, 0.60–0.90).ConclusionsThe right/left adrenal veins diameters ratio was possible predictor of adrenohepatic adhesion.

Baseline Diffusion Weighted Magnetic Resonance Imaging to Predict Recurrence of Anal Squamous Cell Carcinoma (ASCC) Following Chemoradiotherapy

The aim of this study was to investigate the added value of baseline diffusion weighted MRI (DWI) heterogeneity analysis to predict response to chemoradiotherapy (CRT) in patients with primary squamous cell carcinoma of anal canal. We have previously demonstrated that MRI heterogeneity features may have an added value in response prediction in ASCC, here, we present the effect of DWI texture features on disease free survival in an expanded patient cohort. An IRB waiver was obtained for retrospective analysis of standard of care MRI including diffusion weighted sequences from consecutive patients with ASCC undergoing CRT treatment between 2010 and 2016. Whole tumor 3D volume of interest (VOI) was manually delineated on apparent diffusion coefficient (APC) parametric maps with reference to T2 & diffusion weighted imaging by two experienced readers in consensus generating 76 imaging features/VOI. Primary endpoint was disease recurrence (local or metastatic). Patients were censored either at time of event or last follow up. Kaplan Meier survival curves with log rank p-values were calculated for clinic-pathological variables. Most predictive univariate imaging features were selected based on their highest permutation based importance using Random Forest algorithm for survival. 65 patients were included in the current analysis (42 female, 23 male, median age 60 (range 37-85)). 33/65 patients had T3 or T4 disease and 48/65 were node positive. 43 patients achieved a complete response (CR) on re-staging MRI at 10-12 weeks post CRT. Median follow-up was 36 months (2-102). Recurrence occurred in 20/65 patients. Higher T stage (T3/T4), node positivity, gender and presence of CR alone were not significantly associated with improved disease free survival (p=0.52, 0.84, 0.28 and 0.3, respectively). Five baseline DWI imaging features with highest importance were found to be significantly associated with recurrence free survival: ADC maximum, ADC standard deviation, ADC CoV, ADC Log Entropy 1.5, ADC Range (HR>1, p < 0.001). Baseline ADC may have a role in predicting long term outcome in patients with ASCC undergoing CRT treatment and warrants further investigation in prospective trials.