Robust Collaborative Clustering of Subjects and Radiomic Features for Cancer Prognosis.

Feature dimensionality reduction plays an important role in radiomic studies with a large number of features. However, conventional radiomic approaches may suffer from noise, and feature dimensionality reduction techniques are not equipped to utilize latent supervision information of patient data under study, such as differences in patients, to learn discriminative low dimensional representations. To achieve robustness to noise and feature dimensionality reduction with improved discriminative power, we develop a robust collaborative clustering method to simultaneously cluster patients and radiomic features into distinct groups respectively under adaptive sparse regularization. Our method is built upon matrix tri-factorization enhanced by adaptive sparsity regularization for simultaneous feature dimensionality reduction and denoising. Particularly, latent grouping information of patients with distinct radiomic features is learned and utilized as supervision information to guide the feature dimensionality reduction, and noise in radiomic features is adaptively isolated in a Bayesian framework under a general assumption of Laplacian distributions of transform-domain coefficients. Experiments on synthetic data have demonstrated the effectiveness of the proposed approach in data clustering, and evaluation results on an FDG-PET/CT dataset of rectal cancer patients have demonstrated that the proposed method outperforms alternative methods in terms of both patient stratification and prediction of patient clinical outcomes.

COLLABORATIVE CLUSTERING OF SUBJECTS AND RADIOMIC FEATURES FOR PREDICTING CLINICAL OUTCOMES OF RECTAL CANCER PATIENTS.

Most machine learning approaches in radiomics studies ignore the underlying difference of radiomic features computed from heterogeneous groups of patients, and intrinsic correlations of the features are not fully exploited yet. In order to better predict clinical outcomes of cancer patients, we adopt an unsupervised machine learning method to simultaneously stratify cancer patients into distinct risk groups based on their radiomic features and learn low-dimensional representations of the radiomic features for robust prediction of their clinical outcomes. Based on nonnegative matrix tri-factorization techniques, the proposed method applies collaborative clustering to radiomic features of cancer patients to obtain clusters of both the patients and their radiomic features so that patients with distinct imaging patterns are stratified into different risk groups and highly correlated radiomic features are grouped in the same radiomic feature clusters. Experiments on a FDG-PET/CT dataset of rectal cancer patients have demonstrated that the proposed method facilitates better stratification of patients with distinct survival patterns and learning of more effective low-dimensional feature representations that ultimately leads to accurate prediction of patient survival, outperforming conventional methods under comparison.

DEEP CONVOLUTIONAL NEURAL NETWORKS FOR IMAGING DATA BASED SURVIVAL ANALYSIS OF RECTAL CANCER.

Recent radiomic studies have witnessed promising performance of deep learning techniques in learning radiomic features and fusing multimodal imaging data. Most existing deep learning based radiomic studies build predictive models in a setting of pattern classification, not appropriate for survival analysis studies where some data samples have incomplete observations. To improve existing survival analysis techniques whose performance is hinged on imaging features, we propose a deep learning method to build survival regression models by optimizing imaging features with deep convolutional neural networks (CNNs) in a proportional hazards model. To make the CNNs applicable to tumors with varied sizes, a spatial pyramid pooling strategy is adopted. Our method has been validated based on a simulated imaging dataset and a FDG-PET/CT dataset of rectal cancer patients treated for locally advanced rectal cancer. Compared with survival prediction models built upon hand-crafted radiomic features using Cox proportional hazards model and random survival forests, our method achieved competitive prediction performance.

Adaptive Sparsity Regularization Based Collaborative Clustering for Cancer Prognosis.

Radiomic approaches have achieved promising performance in prediction of clinical outcomes of cancer patients. Particularly, feature dimensionality reduction plays an important role in radiomic studies. However, conventional feature dimensionality reduction techniques are not equipped to suppress data noise or utilize latent supervision information of patient data under study (e.g. difference in patients) for learning discriminative low dimensional representations. To achieve feature dimensionality reduction with improved discriminative power and robustness to noisy radiomic features, we develop an adaptive sparsity regularization based collaborative clustering method to simultaneously cluster patients and radiomic features into distinct groups respectively. Our method is built on adaptive sparsity regularized matrix tri-factorization for simultaneous feature denoising and dimension reduction so that the noise is adaptively isolated from the features, and grouping information of patients with distinctive features provides latent supervision information to guide feature dimension reduction. The sparsity regularization is grounded on distribution modeling of transform-domain coefficients in a Bayesian framework. Experiments on synthetic data have demonstrated the effectiveness of the proposed approach in data clustering, and empirical results on an FDG-PET/CT dataset of rectal cancer patients have demonstrated that the proposed method outperforms alternative methods in terms of both patient stratification and prediction of patient clinical outcomes.

A study of positioning orientation effect on segmentation accuracy using convolutional neural networks for rectal cancer.

PURPOSE: Convolutional neural networks (CNN) have greatly improved medical image segmentation. A robust model requires training data can represent the entire dataset. One of the differing characteristics comes from variability in patient positioning (prone or supine) for radiotherapy. In this study, we investigated the effect of position orientation on segmentation using CNN. METHODS: Data of 100 patients (50 in supine and 50 in prone) with rectal cancer were collected for this study. We designed three sets of experiments for comparison: (a) segmentation using the model trained with data from the same orientation; (b) segmentation using the model trained with data from the opposite orientation; (c) segmentation using the model trained with data from both orientations. We performed fivefold cross-validation. The performance was evaluated on segmentation of the clinical target volume (CTV), bladder, and femurs with Dice similarity coefficient (DSC) and Hausdorff distance (HD). RESULTS: Compared with models trained on cases positioned in the same orientation, the models trained with cases positioned in the opposite orientation performed significantly worse (P < 0.05) on CTV and bladder segmentation, but had comparable accuracy for femurs (P > 0.05). The average DSC values were 0.74 vs 0.84, 0.85 vs 0.88, and 0.91 vs 0.91 for CTV, bladder, and femurs, respectively. The corresponding HD values (mm) were 16.6 vs 14.6, 8.4 vs 8.1, and 6.3 vs 6.3, respectively. The models trained with data from both orientations have comparable accuracy (P > 0.05), with average DSC of 0.84, 0.88, and 0.91 and HD of 14.4, 8.1, and 6.3, respectively. CONCLUSIONS: Orientation affects the accuracy for CTV and bladder, but has negligible effect on the femurs. The model trained from data combining both orientations performs as well as a model trained with data from the same orientation for all the organs. These observations can offer guidance on the choice of training data for accurate segmentation.

Cholangiocarcinoma and Gallbladder Cases: An Expert Panel Case-Based Discussion.

Cholangiocarcinoma and gallbladder malignancies are aggressive gastrointestinal malignancies with management dependent on resectability, comorbidities, and location. A multidisciplinary discussion with medical oncologists, radiation oncologists, and surgeons is necessary to determine the optimal treatment approach for each patient. Surgical resection offers the best chance for a long-term cure. Recent studies, such as the phase II SWOG S0809 and the phase III BILCAP study have highlighted the importance of adjuvant treatment with radiation therapy and chemotherapy, respectively, in resected disease. In patients with unresectable disease chemotherapy and chemoradiation therapy to a high dose can improve overall survival and locoregional control. In this expert panel we have brought together radiation oncologists and a medical oncologist to provide case-based feedback on their institutional practices.

Cascaded atrous convolution and spatial pyramid pooling for more accurate tumor target segmentation for rectal cancer radiotherapy.

Convolutional neural networks (CNNs) have become the state-of-the-art method for medical segmentation. However, repeated pooling and striding operations reduce the feature resolution, causing loss of detailed information. Additionally, tumors of different patients are of different sizes. Thus, small tumors may be ignored while big tumors may exceed the receptive fields of convolutions. The purpose of this study is to further improve the segmentation accuracy using a novel CNN (named CAC-SPP) with cascaded atrous convolution (CAC) and a spatial pyramid pooling (SPP) module. This work is the first attempt at applying SPP for segmentation in radiotherapy. We improved the network based on ResNet-101 yielding accuracy gains from a greatly increased depth. We added CAC to extract a high-resolution feature map while maintaining large receptive fields. We also adopted a parallel SPP module with different atrous rates to capture the multi-scale features. The performance was compared with the widely adopted U-Net and ResNet-101 with independent segmentation of rectal tumors for two image sets, separately: (1) 70 T2-weighted MR images and (2) 100 planning CT images. The results show that the proposed CAC-SPP outperformed the U-Net and ResNet-101 for both image sets. The Dice similarity coefficient values of CAC-SPP were 0.78 ± 0.08 and 0.85 ± 0.03, respectively, which were higher than those of U-Net (0.70 ± 0.11 and 0.82 ± 0.04) and ResNet-101 (0.76 ± 0.10 and 0.84 ± 0.03). The segmentation speed of CAC-SPP was comparable with ResNet-101, but about 36% faster than U-Net. In conclusion, the proposed CAC-SPP, which could extract high-resolution features with large receptive fields and capture multi-scale context yields, improves the accuracy of segmentation performance for rectal tumors.

Proton beam reirradiation for locally recurrent pancreatic adenocarcinoma.

BACKGROUND: Local recurrence following definitive treatment for pancreatic adenocarcinoma is common and can be associated with significant morbidity and mortality. Retreatment options for these patients are limited. Proton beam reirradiation (PRT) may limit dose and toxicity to previously irradiated normal tissues in patients without evidence of metastatic disease. METHODS: Between 8/2010-2/2015, 15 patients with isolated, locally-recurrent pancreatic cancer were treated with PRT. Acute toxicity was graded using CTC v 4.0 and defined as occurring within 90 days. Kaplan-Meier survival analysis was performed from the start of PRT. A log-rank test was used to compare survival with or without concurrent chemotherapy. RESULTS: Median follow-up was 15.7 months [2-48] from the start of PRT. The median clinical target volume (CTV) was 71 cc [15-200]. Ten (67%) patients received concurrent chemotherapy. Median PRT dose was 59.4 Gy (37.5-59.4 Gy). The median time interval from the prior treatment course was 26.7 months (7-461.3). There was a rate of 13% acute ≥ grade 3 toxicities attributed to PRT. The median overall survival (OS) was 16.7 months (95% CI, 4.7-36) and OS at 1 year was 67%. The "in-field" failure free survival at one year was 87%. The locoregional progression free survival (LPFS) and distant metastasis free survival (DMFS) at 1 year was 72% and 64% respectively. Concurrent chemotherapy was associated with a higher median survival. CONCLUSIONS: PRT was well tolerated, resulted in prolonged clinical outcomes compared to historical controls, and should be considered as a treatment option with concurrent chemotherapy in selected patients with locally-recurrent pancreatic cancer.

Management of elderly patients with early-stage medically inoperable endometrial cancer: Systematic review and National Cancer Database analysis.

PURPOSE: To evaluate the effectiveness of radiation therapy among elderly patients who are deemed medically inoperable. METHODS AND MATERIALS: We searched PubMed to identify studies from the past 25 years that reported outcomes of medically inoperable endometrial cancer patients treated with radiation alone. The National Cancer Database (NCDB) was queried to identify patients 65 years and older with Stage I-II medically inoperable endometrial cancer. Univariable and multivariable models were performed to investigate the impact of prognostic factors on overall survival. RESULTS: Thirteen papers met inclusion criteria for the systematic review. Overall survival for Stage I tumors at 5 years was 30-95%. Reported pelvic control for the 888 total patients with Stage I tumors was 80-100% and 61-89% for Stage II. Late complications for all patients treated ranged from 0% to 21% across patients. The NCDB analysis demonstrated that any radiotherapy was associated with improved survival over no local therapy. Combination therapy (external beam radiation therapy + brachytherapy) was associated with the most favorable survival with a hazard ratio (HR) of 0.442 (p < 0.001 over no radiotherapy), although benefits were also seen with external beam radiation therapy alone (HR 0.694, p < 0.001) and with brachytherapy alone (HR 0.499, p < 0.001) compared to no radiotherapy. CONCLUSIONS: The available evidence suggests high rates of local control after radiation therapy for elderly women with Stage I-II medically inoperable endometrial cancer. Our analysis of the NCDB suggests that radiation therapy improves survival, and combination therapy provides the most favorable outcomes. Given a relatively favorable toxicity profile, definitive radiation therapy should be considered a preferred approach for patients with medically inoperable endometrial cancer.

Contribution of CXCL12 secretion to invasion of breast cancer cells.

INTRODUCTION: Neu (HER2/ErbB2) is overexpressed in 25% to 30% of human breast cancer, correlating with a poor prognosis. Researchers in previous studies who used the mouse mammary tumor virus Neu-transgenic mouse model (MMTV-Neu) demonstrated that the Neu-YB line had increased production of CXCL12 and increased metastasis, whereas the Neu-YD line had decreased metastasis. In this study, we examined the role of increased production of CXCL12 in tumor cell invasion and malignancy. METHODS: We studied invasion in the tumor microenvironment using multiphoton intravital imaging, in vivo invasion and intravasation assays. CXCL12 signaling was altered by using the CXCR4 inhibitor AMD3100 or by increasing CXCL12 expression. The role of macrophage signaling in vivo was determined using a colony-stimulating factor 1 receptor (CSF-1R) blocking antibody. RESULTS: The Neu-YD strain was reduced in invasion, intravasation and metastasis compared to the Neu-YB and Neu deletion mutant (activated receptor) strains. Remarkably, in the Neu-YB strain, in vivo invasion to epidermal growth factor was dependent on both CXCL12-CXCR4 and CSF1-CSF-1R signaling. Neu-YB tumors had increased macrophage and microvessel density. Overexpression of CXCL12 in rat mammary adenocarcinoma cells increased in vivo invasion as well as microvessel and macrophage density. CONCLUSIONS: Expression of CXCL12 by tumor cells results in increased macrophage and microvessel density and in vivo invasiveness.

Tumor-stroma: In vivo assays and intravital imaging to study cell migration and metastasis.

The development of metastatic disease is often correlated with poor patient outcome in a variety of different cancers. The metastatic cascade is a complex, multistep process that involves the growth of the primary tumor and angiogenesis, invasion into the local environment, intravasation into the vasculature, tumor cell survival in the circulation, extravasation from the vasculature and sustained growth at secondary organ sites to form metastases. Although in vitro assays of single cell types can provide information regarding cell autonomous mechanisms contributing to metastasis, the in vivo microenvironment entails a network of interactions between cells which is also important. Insight into the mechanisms underlying tumor cell migration, invasion and metastasis in vivo has been aided by development of multiphoton microscopy and in vivo assays, which we will review here.

A functional in vivo screen for regulators of tumor progression identifies HOXB2 as a regulator of tumor growth in breast cancer.

Microarray profiling in breast cancer patients has identified genes correlated with prognosis whose functions are unknown. The purpose of this study was to develop an in vivo assay for functionally screening regulators of tumor progression using a mouse model. Transductant shRNA cell lines were made in the MDA-MB-231 breast cancer line. A pooled population of 25 transductants was injected into the mammary fat pads and tail veins of mice to evaluate tumor growth, and experimental metastasis. The proportions of transductants were evaluated in the tumor and metastases using barcodes specific to each shRNA transductant. We characterized the homeobox 2 transcription factor as a negative regulator, decreasing tumor growth in MDA-MB-231, T47D, and MTLn3 mammary adenocarcinoma cell lines. Homeobox genes have been correlated with cancer patient prognosis and tumorigenesis. Here we use a novel in vivo shRNA screen to identify a new role for a homeobox gene in human mammary adenocarcinoma.

ERBB1 and ERBB2 have distinct functions in tumor cell invasion and intravasation.

PURPOSE: The epidermal growth factor receptor (ERBB1) and related family member HER-2/neu (ERBB2) are often overexpressed in aggressive breast cancers and their overexpression is correlated with poor prognosis. Clinical studies using ERBB inhibitors have focused on tumor growth effects, but ERBBs can contribute to malignancy independent of their effects on tumor growth. Our studies were designed to evaluate the effect of ERBB inhibition on tumor cell motility and intravasation in vivo using clinically relevant small-molecule inhibitors. EXPERIMENTAL DESIGN: Using in vivo mouse models of breast cancer, we test the effects of ERBB1 and ERBB2 inhibitors AC480 and lapatinib, ERBB1 inhibitor gefitinib, and ERBB2 inhibitor AG825 on in vivo tumor cell invasive properties in mammary fat pad tumors. RESULTS: ERBB1 and ERBB2 inhibition rapidly (within 3 h) inhibits both tumor cell motility and intravasation. Using gefitinib, ERBB1 inhibition rapidly inhibits tumor cell motility and invasion but not intravasation, whereas ERBB2 inhibition by AG825 rapidly blocks intravasation. CONCLUSIONS: ERBB1 and ERBB2 inhibition can rapidly block tumor cell invasive properties. In addition, we differentiate for the first time the contributions of ERBB1 and ERBB2 to the key metastatic properties of in vivo tumor cell invasion and intravasation. These experiments temporally and molecularly separate two key stages in tumor cell entry into blood vessels: invasion and intravasation. These results indicate that ERBB inhibition should be considered for blocking other tumor cell malignant properties besides growth.

Identification of cellular proteins that maintain retroviral epigenetic silencing: evidence for an antiviral response.

Integrated retroviral DNA is subject to epigenetic gene silencing, resulting in loss of expression of viral genes as well as reporter or therapeutic genes transduced by retroviral vectors. Possible mediators of such silencing include the histone deacetylase (HDAC) family of cellular proteins. We previously isolated HeLa cell populations that harbored silent avian sarcoma virus-based green fluorescent protein (GFP) vectors that could be reactivated by treatment with HDAC inhibitors. Here, we developed a small interfering RNA (siRNA)-based approach to identify specific host factors that participate in the maintenance of silencing. Knockdown of HDAC1, the transcriptional repressor Daxx (a binding partner of HDAC1), or heterochromatin protein 1 gamma resulted in robust and specific GFP reporter gene reactivation. Analyses of cell clones and diverse GFP vector constructs revealed that the roles of HDAC1 and Daxx in retroviral silencing are largely independent of the integration site or the promoter controlling the silent GFP reporter gene. Previous findings from our laboratory and those of others have suggested that Daxx and HDAC proteins may act broadly as part of an antiviral response to repress viral gene transcription. Expression of presumptive viral "countermeasure" proteins that are known to inhibit Daxx or HDACs (pp71, IE2, and Gam1) resulted in the reactivation of GFP reporter gene expression. This study has identified individual host factors that maintain retroviral silencing and supports the proposal that these factors participate in an antiviral response. Furthermore, our results indicate that siRNAs can be used as specific reagents to interrupt the maintenance of epigenetic silencing.

High-frequency epigenetic repression and silencing of retroviruses can be antagonized by histone deacetylase inhibitors and transcriptional activators, but uniform reactivation in cell clones is restricted by additional mechanisms.

Integrated retroviral DNA is subject to epigenetic gene silencing, but the viral and host cell properties that influence initiation, maintenance, and reactivation are not fully understood. Here we describe rapid and high-frequency epigenetic repression and silencing of integrated avian sarcoma virus (ASV)-based vector DNAs in human HeLa cells. Initial studies utilized a vector carrying the strong human cytomegalovirus (hCMV) immediate-early (IE) promoter to drive expression of a green fluorescent protein (GFP) reporter gene, and cells were sorted into two populations based on GFP expression [GFP(+) and GFP(-)]. Two potent epigenetic effects were observed: (i) a very broad distribution of GFP intensities among cells in the GFP(+) population as well as individual GFP(+) clones and (ii) high-frequency GFP reporter gene silencing in GFP(-) cells. We previously showed that histone deacetylases (HDACs) can associate with ASV DNA soon after infection and may act to repress viral transcription at the level of chromatin. Consistent with this finding, we report here that treatment with the histone deacetylase inhibitor trichostatin A (TSA) induces GFP activation in GFP(-) cells and can also increase GFP expression in GFP(+) cells. In the case of the GFP(-) populations, we found that after removal of TSA, GFP silencing was reestablished in a subset of cells. We used that finding to enrich for stable GFP(-) cell populations in which viral GFP reporter expression could be reactivated by TSA; furthermore, we found that the ability to isolate such populations was independent of the promoter driving the GFP gene. In such enriched cultures, hCMV IE-driven, but not the viral long terminal repeat-driven, silent GFP reporter expression could be reactivated by the transcriptional activator prostratin. Microscopy-based studies using synchronized cells revealed variegated reactivation in cell clones, indicating that secondary epigenetic effects can restrict reactivation from silencing. Furthermore we found that entry into S phase was not required for reactivation. We conclude that HDACs can act rapidly to initiate and maintain promoter-independent retroviral epigenetic repression and silencing but that reactivation can be restricted by additional mechanisms.

The cloning and characterization of the histone acetyltransferase human homolog Dmel\TIP60 in Drosophila melanogaster: Dmel\TIP60 is essential for multicellular development.

Chromatin packaging directly influences gene programming as it permits only certain portions of the genome to be activated in any given developmental stage, cell, and tissue type. Histone acetyltransferases (HATs) are a key class of chromatin regulatory proteins that mediate such developmental chromatin control; however, their specific roles during multicellular development remain unclear. Here, we report the first isolation and developmental characterization of a Drosophila HAT gene (Dmel\TIP60) that is the homolog of the human HAT gene TIP60. We show that Dmel\TIP60 is differentially expressed during Drosophila development, with transcript levels significantly peaking during embryogenesis. We further demonstrate that reducing endogenous Dmel\TIP60 expression in Drosophila embryonic cells by RNAi results in cellular defects and lethality. Finally, using a GAL4-targeted RNAi system in Drosophila, we show that ubiquitous or mesoderm/muscle-specific reduction of Dmel\TIP60 expression results in lethality during fly development. Our results suggest a mechanism for HAT regulation involving developmental control of HAT expression profiles and show that Dmel\TIP60 is essential for multicellular development. Significantly, our inducible and targeted HAT knockdown system in Drosophila now provides a powerful tool for effectively studying the roles of TIP60 in specific tissues and cell types during development.

Human immunodeficiency virus type 1 DNA nuclear import and integration are mitosis independent in cycling cells.

An essential step in human immunodeficiency virus type 1 (HIV-1) replication is the movement of the viral preintegration complex from the cytoplasm into the nucleus. The pathway(s) and timing for HIV-1 DNA nuclear entry in cycling cells have not been established. Here, we show that if cycling cells are infected before S phase, viral DNA can be integrated prior to passage of the host DNA replication fork through the integration site, as indicated by stable inheritance in both daughter cells. We conclude that efficient nuclear entry can occur independently of mitotic nuclear disassembly in cycling cells.

Transduction of interphase cells by avian sarcoma virus.

It has been generally believed that oncoretroviruses are dependent on mitosis for efficient nuclear entry of viral DNA. We previously identified a nuclear localization signal in the integrase protein of an oncoretrovirus, avian sarcoma virus (ASV), suggesting an active import mechanism for the integrase-DNA complex (G. Kukolj, R. A. Katz, and A. M. Skalka, Gene 223:157-163, 1998). Here, we have evaluated the requirement for mitosis in nuclear import and integration of ASV DNA. Using a modified ASV encoding a murine leukemia virus amphotropic env gene and a green fluorescent protein (GFP) reporter gene, DNA nuclear import was measured in cell cycle-arrested avian (DF-1) as well as human (HeLa) and mouse cells. The results showed efficient accumulation of nuclear forms of ASV DNA in gamma-irradiation-arrested cells. Efficient transduction of a GFP reporter gene was also observed after infection of cells that were arrested with gamma-irradiation, mitomycin C, nocodazole, or aphidicolin, confirming that nuclear import and integration of ASV DNA can occur in the absence of mitosis. By monitoring GFP expression in individual cells, we also obtained evidence for nuclear import of viral DNA during interphase in cycling cells. Lastly, we observed that ASV can transduce postmitotic mouse neurons. These results support an active nuclear import mechanism for the oncoretrovirus ASV and suggest that this mechanism can operate in both nondividing and dividing cells.