A co-design framework of neural networks and quantum circuits towards quantum advantage.

Despite the pursuit of quantum advantages in various applications, the power of quantum computers in executing neural network has mostly remained unknown, primarily due to a missing tool that effectively designs a neural network suitable for quantum circuit. Here, we present a neural network and quantum circuit co-design framework, namely QuantumFlow, to address the issue. In QuantumFlow, we represent data as unitary matrices to exploit quantum power by encoding n = 2k inputs into k qubits and representing data as random variables to seamlessly connect layers without measurement. Coupled with a novel algorithm, the cost complexity of the unitary matrices-based neural computation can be reduced from O(n) in classical computing to O(polylog(n)) in quantum computing. Results show that on MNIST dataset, QuantumFlow can achieve an accuracy of 94.09% with a cost reduction of 10.85 × against the classical computer. All these results demonstrate the potential for QuantumFlow to achieve the quantum advantage.

On Interpretability of Artificial Neural Networks: A Survey.

Deep learning as represented by the artificial deep neural networks (DNNs) has achieved great success recently in many important areas that deal with text, images, videos, graphs, and so on. However, the black-box nature of DNNs has become one of the primary obstacles for their wide adoption in mission-critical applications such as medical diagnosis and therapy. Because of the huge potentials of deep learning, increasing the interpretability of deep neural networks has recently attracted much research attention. In this paper, we propose a simple but comprehensive taxonomy for interpretability, systematically review recent studies in improving interpretability of neural networks, describe applications of interpretability in medicine, and discuss possible future research directions of interpretability, such as in relation to fuzzy logic and brain science.

Improved Linear Convergence of Training CNNs With Generalizability Guarantees: A One-Hidden-Layer Case.

We analyze the learning problem of one-hidden-layer nonoverlapping convolutional neural networks with the rectified linear unit (ReLU) activation function from the perspective of model estimation. The training outputs are assumed to be generated by the neural network with the unknown ground-truth parameters plus some additive noise, and the objective is to estimate the model parameters by minimizing a nonconvex squared loss function of the training data. Assuming that the training set contains a finite number of samples generated from the Gaussian distribution, we prove that the accelerated gradient descent (GD) algorithm with a proper initialization converges to the ground-truth parameters (up to the noise level) with a linear rate even though the learning problem is nonconvex. Moreover, the convergence rate is proved to be faster than the vanilla GD. The initialization can be achieved by the existing tensor initialization method. In contrast to the existing works that assume an infinite number of samples, we theoretically establish the sample complexity of the required number of training samples. Although the neural network considered here is not deep, this is the first work to show that accelerated GD algorithms can find the global optimizer of the nonconvex learning problem of neural networks. This is also the first work that characterizes the sample complexity of gradient-based methods in learning convolutional neural networks with the nonsmooth ReLU activation function. This work also provides the tightest bound so far of the estimation error with respect to the output noise.

Universal approximation with quadratic deep networks.

Recently, deep learning has achieved huge successes in many important applications. In our previous studies, we proposed quadratic/second-order neurons and deep quadratic neural networks. In a quadratic neuron, the inner product of a vector of data and the corresponding weights in a conventional neuron is replaced with a quadratic function. The resultant quadratic neuron enjoys an enhanced expressive capability over the conventional neuron. However, how quadratic neurons improve the expressing capability of a deep quadratic network has not been studied up to now, preferably in relation to that of a conventional neural network. Specifically, we ask four basic questions in this paper: (1) for the one-hidden-layer network structure, is there any function that a quadratic network can approximate much more efficiently than a conventional network? (2) for the same multi-layer network structure, is there any function that can be expressed by a quadratic network but cannot be expressed with conventional neurons in the same structure? (3) Does a quadratic network give a new insight into universal approximation? (4) To approximate the same class of functions with the same error bound, could a quantized quadratic network have a lower number of weights than a quantized conventional network? Our main contributions are the four interconnected theorems shedding light upon these four questions and demonstrating the merits of a quadratic network in terms of expressive efficiency, unique capability, compact architecture and computational capacity respectively.

ColXα1 is a stromal component that colocalizes with elastin in the breast tumor extracellular matrix.

The tumor microenvironment regulates tissue development and homeostasis, and its dysregulation contributes to neoplastic progression. Increased expression of type X collagen α-1 (ColXα1) in tumor-associated stroma correlates with poor pathologic response to neoadjuvant chemotherapy in estrogen receptor (ER) and human epidermal growth factor receptor 2 (HER2)-positive breast cancers. Evaluation of ColXα1 expression patterns suggests a potential connection with elastin fibers. To investigate the possible interaction between ColXα1 and elastin, we evaluated the expression of ColXα1 in relation to elastin fibers in normal breast tissue, ductal carcinoma in situ, and invasive breast carcinomas at cellular and subcellular levels. Our findings demonstrate that ColXα1 colocalizes with elastin in invasive breast cancer-associated stroma by immunohistochemistry, immunofluorescence, and electron microscopy. In 212 invasive breast carcinomas, this complex was aberrantly and selectively expressed in tumor extracellular matrix in 79% of ER+/HER2-, 80% of ER+/HER2+, 76% of ER-/HER2+, and 58% of triple negative breast cancers. In contrast, ColXα1 was generally absent, while elastin was present perivascularly in normal breast tissue. ColXα1 and elastin were coexpressed in 58% of ductal carcinoma in situ (DCIS) in periductal areas. In mass-forming DCIS with desmoplastic stroma, the complex was intensely expressed in periductal areas as well as within the tumor-associated stroma in all cases. Our data suggest that the breast carcinoma neoplastic process may involve aberrant expression of ColXα1 and elastin in the tumor microenvironment emerging early at the DCIS stage. Enrichment of these complexes in tumor-associated stroma may represent a stromal signature indicative of intrinsic differences between breast cancers. These findings shed light on investigation into the role of aberrant collagen complex expression in tumorigenesis and tumor progression which may be leveraged in therapeutic and theranostic applications.

Endometrial Adenocarcinomas With Significant Mucinous Differentiation: A Characterization of Intratumoral Heterogeneity of KRAS Mutations in Mucinous and Endometrioid Histologic Components.

OBJECTIVE: KRAS mutations are frequently seen in malignancies with mucinous morphology. In our previous study, mucinous endometrial carcinomas were associated with a significantly higher frequency of KRAS mutations as compared with matched conventional endometrioid carcinomas. This study expands our previous report by exploring possible intratumoral heterogeneity for KRAS gene mutations in the mucinous components of mucinous carcinomas (MCs) and endometrioid carcinomas with significant mucinous differentiation (ECSMD) versus their associated "usual" endometrioid components. MATERIALS AND METHODS: KRAS-positive cases from our previous report were studied, including 10 MCs and 10 ECSMDs. The specimens were microscopically dissected to separately isolate morphologically mucinous and endometrioid components. Direct DNA sequencing for KRAS mutations at codons 12 and 13 using capillary electrophoresis were performed. RESULTS: KRAS mutations were detected in the endometrioid components of 8 (80%) of 10 MCs and 3 (30%) of 10 ECSMDs. The endometrioid component of the ECSMD group was less frequently associated with KRAS mutation than the endometrioid component of the MC group, even when the mucinous component of the same tumor contained a mutation; the difference is statistically significant (P < 0.05). CONCLUSIONS: Our current study shows that intratumoral heterogeneity for KRAS gene mutation was associated with ECSMD, but less frequently with MC. It is possible that when the mucinous component predominates, qualifying for an MC, KRAS mutations appear to be widespread, irrespective of the mucinous or nonmucinous differentiation of the tumor cells. The findings suggest that multiple samples for KRAS tests may be useful, especially in endometrioid carcinoma with significant mucinous differentiation.

Stromal Clusterin Expression Predicts Therapeutic Response to Neoadjuvant Chemotherapy in Triple Negative Breast Cancer.

BACKGROUND: Expression of clusterin correlates with tumor progression and therapeutic response in several human malignancies, including breast cancer. However, its predictive value in the neoadjuvant setting in breast cancer remains unexplored. The objective of this explorative study was to determine whether clusterin expression in breast cancer correlated with clinical pathologic characteristics and whether its expression was predictive of response to neoadjuvant chemotherapy (NAC). MATERIALS AND METHODS: We determined the clusterin expression pattern in 72 triple negative breast cancers (TNBC) treated with NAC before surgery. Clusterin expression was evaluated by immunohistochemistry and was correlated with pathologic characteristics and response to NAC using residual cancer burden score. RESULTS: Immunohistochemistry analysis revealed a differential pattern of expression between tumor and stroma. Clusterin expression in the tumor associated stroma as opposed to expression by the neoplastic epithelium was significantly associated with neoadjuvant-treated TNBC. Low stromal clusterin, low stromal content, and high tumor-infiltrating lymphocytes were associated with a significantly greater likelihood of achieving a good pathologic response as reflected by lower residual cancer burden scores (P = .002, P = .003, and P = .001, respectively). Tumor and/or stromal clusterin expression were not associated with patient age, tumor histologic grade, stage, and lymph node status. CONCULSION: This study suggests a potential role for the assessment of stromal clusterin as a predictive biomarker for response of TNBC to neoadjuvant therapy. Further validation of this biomarker in a large study is needed.

The clinical significance of K-ras mutation in endometrial "surface epithelial changes" and their associated endometrial adenocarcinoma.

OBJECTIVES: The entity of 'surface epithelial changes' (SECs) was first described in 1995 [1]. Morphologically, SECs usually arise from malignant glands at the superficial aspect of well differentiated (WD) endometrioid carcinomas (ECs) and impart the appearance of a 'maturational' phenomenon at the surface of the cancer. Exhibiting a paradoxically bland histologic appearance, SECs typically show morphologic features that mimic benign entities, particularly endocervical microglandular hyperplasia (MGH). SECs have been associated with approximately half of WD endometrioid carcinomas many of which showed focal mucinous differentiation. Despite their morphologically benign histology, some have questioned whether the presence of SECs represents a 'marker' for an underlying malignancy, especially in postmenopausal women with endocervical or MGH-type SECs in their endometrial sampling. Since the biologic nature of SECs is unknown, we aimed to study the prevalence of KRAS gene mutations in SECs and the underlying WD endometrioid adenocarcinomas (EC) from which they directly arise. METHODS: 24 cases with biopsy proven SECs and ECs in their subsequent hysterectomy were retrieved. Genomic DNA was extracted from formalin-fixed paraffin-embedded tissue. PCR amplification for KRAS codons 12 and 13 was performed, followed by sequencing using capillary electrophoresis. RESULTS: KRAS codons 12 and 13 mutations were detected in 19 of 24 (79%) SECs, and 19 of 24 (79%) ECs. All SECs had the same KRAS mutation as the underlying EC. CONCLUSIONS: Our results suggest that SECs are of neoplastic origin and that KRAS mutations play an important role in the tumorigenesis of ECs and SECs.

Identification of stromal ColXα1 and tumor-infiltrating lymphocytes as putative predictive markers of neoadjuvant therapy in estrogen receptor-positive/HER2-positive breast cancer.

BACKGROUND: The influence of the tumor microenvironment and tumor-stromal interactions on the heterogeneity of response within breast cancer subtypes have just begun to be explored. This study focuses on patients with estrogen receptor-positive/human epidermal growth factor receptor 2-positive (ER+/HER2+) breast cancer receiving neoadjuvant chemotherapy and HER2-targeted therapy (NAC+H), and was designed to identify novel predictive biomarkers by combining gene expression analysis and immunohistochemistry with pathologic response. METHODS: We performed gene expression profiling on pre-NAC+H tumor samples from responding (no or minimal residual disease at surgery) and non-responding patients. Gene set enrichment analysis identified potentially relevant pathways, and immunohistochemical staining of pre-treatment biopsies was used to measure protein levels of those pathways, which were correlated with pathologic response in both univariate and multivariate analysis. RESULTS: Increased expression of genes encoding for stromal collagens, including Col10A1, and reduced expression of immune-associated genes, reflecting lower levels of total tumor-infiltrating lymphocytes (TILs), were strongly associated with poor pathologic response. Lower TILs in tumor biopsies correlated with reduced likelihood of achieving an optimal pathologic response, but increased expression of the Col10A1 gene product, colXα1, had greater predictive value than stromal abundance for poor response (OR = 18.9, p = 0.003), and the combination of increased colXα1 expression and low TILs was significantly associated with poor response in multivariate analysis. ROC analysis suggests strong specificity and sensitivity for this combination in predicting treatment response. CONCLUSIONS: Increased expression of stromal colXα1 and low TILs correlate with poor pathologic response in ER+/HER2+ breast tumors. Further studies are needed to confirm their predictive value and impact on long-term outcomes, and to determine whether this collagen exerts a protective effect on the cancer cells or simply reflects other factors within the tumor microenvironment.

KRAS Mutations in Mucinous Lesions of the Uterus.

OBJECTIVES: The current study examined the KRAS mutation status in a spectrum of mucinous lesions of the uterus, including mucinous metaplasia (MM), atypical mucinous proliferation (AMP), endocervical mucosa, and microglandular hyperplasia (MGH). METHODS: Thirty-nine cases, including 15 AMPs, nine MMs, nine MGHs, and six normal endocervical mucosas, were selected from the departmental archive. All AMP cases with follow-up biopsies or hysterectomies were reviewed. Genomic DNA was extracted from formalin-fixed, paraffin-embedded tissue and KRAS codons 12 and 13 sequence analyzed. RESULTS: KRAS codon 12 and 13 mutations were detected in 10 (67%) of the 15 AMP cases. No KRAS mutations were identified in MMs, MGHs, and endocervical mucosas (P = .002, AMP vs MM or MGH, Fisher exact test). Most women with AMP were postmenopausal (13/15 [86.7%]) and presented with dysfunctional uterine bleeding. Among the 10 cases of AMP harboring KRAS mutations, six (60%) cases were subsequently diagnosed with carcinoma, one with atypical complex hyperplasia, and two with AMP within endometrial polyps. CONCLUSIONS: The results suggest a possible association between KRAS mutations and mucinous differentiation in endometrial carcinogenesis. KRAS status can help in assessing benign from precursor or malignant mucinous lesions as well as differentiate endometrial lesions from those of cervical origin.

Endometrial carcinomas with significant mucinous differentiation associated with higher frequency of k-ras mutations: a morphologic and molecular correlation study.

OBJECTIVES: K-ras gene product in the mitogen-activated protein kinase/extracellular signal-regulated kinase pathway is critical in the development of certain types of malignancies. K-ras mutation-associated pancreatic and ovarian carcinomas often display mucinous differentiation. Previous studies have shown that k-ras mutation is found in 10% to 30% of endometrial carcinomas. We investigated k-ras mutations in several morphologic subtypes of endometrial carcinomas with particular emphasis on various degrees of mucinous differentiation. METHODS: Genomic DNA was extracted from formalin-fixed paraffin-embedded (FFPE) tissue sections. Polymerase chain reaction amplification for k-ras codons 12 and 13 were performed, followed by sequencing using capillary electrophoresis. The Fisher exact test is used to compare the prevalent difference of k-ras mutation among the groups. P < 0.05 was considered significant. RESULTS: K-ras mutations were detected in 8 (80%) of 10 mucinous carcinomas, 12 (67%) of 18 endometrioid carcinomas (ECs) with significant mucinous differentiation (ECMD), 4 (25%) of 16 ECs, and 1 (9%) of 11 serous carcinomas. The differences were statistically significant between mucinous carcinomas versus EC (P < 0.01) and ECMD versus EC (P < 0.05). CONCLUSION: The findings suggest that mucinous carcinoma and endometrioid carcinoma with significant mucinous component are more likely to be associated with k-ras mutation. Potential clinical implications of k-ras mutation lies in the management of recurrent or higher-stage endometrial mucinous tumors, which would not be responsive to treatment protocols containing epidermal growth factor receptor inhibitors.

BRAF V600E Mutations in Endometrial Adenocarcinoma.

The BRAF V600E somatic mutation is recognized as an oncogenic driver of many human cancers involving the MAPK/ERK pathway. Colorectal and lung cancers associated with the BRAF V600E mutation often demonstrated mucinous morphology. This study hypothesized that the BRAF V600E mutation may be associated with mucinous morphology in endometrial cancer and aimed to investigate its prevalence in mucinous (endometrial) carcinoma (MC) and endometrioid adenocarcinoma with significant mucinous differentiations (ECMD) (>10% neoplastic cells). Twenty-eight cases of endometrial cancer were selected, including 17 (60.7%) cases of MC or ECMD. All patients were Caucasian with age ranging from 50 to 87 years old (median 65). Genomic DNA was extracted from formalin-fixed paraffin-embedded tumor tissue and subjected to both real-time mutant allele-specific amplification polymerase chain reaction (PCR) and PCR amplification, followed by direct sequencing. Three (3/28, 10.7%) BRAF V600E mutations were detected by real-time mutant allele-specific amplification PCR and confirmed by direct sequencing. Two of 3 cases positive for BRAF V600E mutation were ECMDs with "surface epithelial changes." KRAS mutations were found in 9 cases (32.1%), none with BRAF mutation. This is the first report of BRAF V600E mutation in endometrial cancer, indicating that it may contribute to tumorigenesis of endometrial cancer, although at a low frequency compared with KRAS mutations.

A Rapidly Enlarging Squamous Inclusion Cyst in an Axillary Lymph Node following Core Needle Biopsy.

A 73-year-old woman was found to have a 1.7 cm axillary mass, for which a core needle biopsy was performed. The specimen revealed fragmented squamous epithelium surrounded by lymphoid tissue consistent with a squamous inclusion cyst in a lymph node, but a metastatic squamous cell carcinoma could not be excluded. Within one month, the lesion enlarged to 5 cm and was excised. Touch preparation cytology during intraoperative consultation displayed numerous single and sheets of atypical epithelioid cells with enlarged nuclei and occasional mitoses, suggesting a carcinoma. However, multinucleated giant cells and neutrophils in the background indicated reactive changes. We interpreted the touch preparation as atypical and recommended conservative surgical management. Permanent sections revealed a ruptured squamous inclusion cyst in a lymph node with extensive reactive changes. Retrospectively, the atypical epithelioid cells on touch preparation corresponded to reactive histiocytes. This is the first case report of a rapidly enlarging ruptured squamous inclusion cyst in an axillary lymph node following core needle biopsy. Our case demonstrates the diagnostic challenges related to a ruptured squamous inclusion cyst and serves to inform the readers to consider this lesion in the differential diagnosis for similar situations.

[Ecological control effects of Litchi chinensis-Desmodium intortum complex plant ecosystem on litchi pests].

An investigation on the community structure and dynamics of litchi pests and their natural enemies in constructed Litchi chinensis-Desmodium intortum complex plant ecosystem and single L. chinensis ecosystem showed that the total amount of litchi pests in the complex plant ecosystem was 61.27% of that in the single ecosystem in whole year, and only 50.45% in May, the key time for fruit development, which suggested that there was an interaction between D. intortum and L. chinensis. D. intortum and L. chinensis had a few common pests, but many common natural enemies. D. intortum florescence in winter provided shelter and substitutive food for the natural enemies of pests to survive in the extreme environmental conditions in winter. L. chinensis florescence was on the heel of D. intortum florescence, which provided better conditions for the natural enemies to survive and multiply. During florescence and fruit development stages of L. chinensis (from March to June), the predator/prey ratio in complex plant system was 4.22, 2.34, 2.2 and 20.63 times of that in single plant system in March, April, May and June, respectively, indicating the good control effect on pests of L. chinensis.

[Diversity and stability of arthropod assemblage in tea orchard].

Two tea orchards, simplex tea orchard with weeds removed manually or by herbicides (STO) and complex tea orchard with the weed Hedyotis uncinella (CTO), each with an area of 0. 4 hm2, were established in 1995 in Yingde Hongxing Tea Plantation, Guangdong Province. The primary eigenvalues, species richness index (R), assemblage diversity index (H'), evenness index (J) and species concentration index (C) of arthropod assemblage were employed and compared to assess the efficacy of STO and CTO on the diversity and stabilityof arthropod assemblage. Stability indexes Ss/Si and Sn/Sp and variation coefficient of diversity index ds/dm were utilized as well. The results demonstrated that the R of arthropod assemblage in CTO ranged from 4 to 8, with the highest of 7.7403, while that in STO varied mainly between 4 to 6. The average R of arthropod assemblage in CTO was 5.4672 +/- 0.3483, higher than that in STO (4.8809 +/- 0.3175). The H' of arthropod in CTO (3.8535 +/- 0.1232) was higher, in contrast to the value in STO (3.4654 +/- 0.1856). The J in CTO was higher, while the species concentration index (C) was lower, in comparison to STO. The stability indexes Ss/Si and Sn/Sp of CTO were greater than those of STO, while the ds/dm in CTO (0.1107) was lower than that in STO (0.1855). All these indicated that the diversity of arthropod assemblage was better preserved in CTO, and the assemblage in CTO was more stable.

Effects of load and speed on lumbar vertebral kinematics during lifting motions.

This experimental study investigated the effects of load magnitude and movement speed on lumbar vertebral kinematics during lifting task performance. Ten participants performed sagittally symmetric lifting movements with systematically varied load using either a normal or a faster-than-normal speed. Skin-surface markers were strategically placed over the participants' spinous processes and other landmarks representing major body joints and were recorded during the movements by a motion capture system. The center of rotation (COR) locations and segmental movement profiles for lumbar vertebrae L2 to L5 were derived and analyzed. Results suggested that (a) the COR locations and vertebral angular displacement were not significantly affected by the speed or load variation; (b) a faster speed tended to shorten the time to complete the acceleration for all the lumbar vertebrae considered; and (c) the load increase incurred a tendency for the L5 to complete the primary displacement in a briefer time while enduring greater peak acceleration and velocity. The findings lead to a better understanding of the relation between lifting dynamics and spinal motion. Potential applications of this research include the development of more accurate biomechanical models and software tools for depicting spinal motions and quantifying low-back stress.

Model-guided derivation of lumbar vertebral kinematics in vivo reveals the difference between external marker-defined and internal segmental rotations.

This study investigated whether the external marker-defined spine inter-segmental rotation is different from the internal vertebral rotation, and explored how to estimate the latter from limited surface measurement. A kinematic model was first created to elucidate analytically the relation between the external and internal rotations. A novel approach guided by the model was proposed for deriving vertebral centers of rotation (CORs) from measured planar trajectories of skin-surface markers. The approach involved a recursive procedure for establishing local (anatomical) coordinate systems, and an optimization routine that identified the maximum-likelihood circles best fitting the marker trajectories in local coordinate systems. An experiment with 10 subjects (5 males and 5 females) was conducted to test the approach along with the model. Skin-surface markers were strategically placed over individual spinous processes and other body landmarks, and recorded by an opto-electronic system while sagittally symmetric load-lifting movements were being performed. For the majority (89%) of measured motions, the COR locations for lumbar vertebrae (L2-L5) were derived successfully: solutions resulting from the optimization routine met a convergence criterion governed by the model, and were in agreement with existing data from radiographic or cadaveric studies. Empirical results confirmed the differences between the external marker-defined inter-segmental motions and corresponding internal vertebral rotations (1.1-5.8 degrees on average, all statistically significant). The study demonstrated the necessity and viability of quantifying internal vertebral kinematics when utilizing non-invasive marker-based measurement for spine-related clinical diagnosis and biomechanical modeling.