Learning Methods and Algorithms for Semantic Text Classification across Multiple Domains

Information is nowadays a key resource: machine learning and data mining techniques have been developed to extract high-level information from great amounts of data. As most data comes in form of unstructured text in natural languages, research on text mining is currently very active and dealing with practical problems. Among these, text categorization deals with the automatic organization of large quantities of documents in priorly defined taxonomies of topic categories, possibly arranged in large hierarchies. In commonly proposed machine learning approaches, classifiers are automatically trained from pre-labeled documents: they can perform very accurate classification, but often require a consistent training set and notable computational effort. Methods for cross-domain text categorization have been proposed, allowing to leverage a set of labeled documents of one domain to classify those of another one. Most methods use advanced statistical techniques, usually involving tuning of parameters. A first contribution presented here is a method based on nearest centroid classification, where profiles of categories are generated from the known domain and then iteratively adapted to the unknown one. Despite being conceptually simple and having easily tuned parameters, this method achieves state-of-the-art accuracy in most benchmark datasets with fast running times. A second, deeper contribution involves the design of a domain-independent model to distinguish the degree and type of relatedness between arbitrary documents and topics, inferred from the different types of semantic relationships between respective representative words, identified by specific search algorithms. The application of this model is tested on both flat and hierarchical text categorization, where it potentially allows the efficient addition of new categories during classification. Results show that classification accuracy still requires improvements, but models generated from one domain are shown to be effectively able to be reused in a different one.

Segmentazione e categorizzazione di oggetti mediante immagini depth e deep learning

Questo lavoro è iniziato con uno studio teorico delle principali tecniche di classificazione di immagini note in letteratura, con particolare attenzione ai più diffusi modelli di rappresentazione dell’immagine, quali il modello Bag of Visual Words, e ai principali strumenti di Apprendimento Automatico (Machine Learning). In seguito si è focalizzata l’attenzione sulla analisi di ciò che costituisce lo stato dell’arte per la classificazione delle immagini, ovvero il Deep Learning. Per sperimentare i vantaggi dell’insieme di metodologie di Image Classification, si è fatto uso di Torch7, un framework di calcolo numerico, utilizzabile mediante il linguaggio di scripting Lua, open source, con ampio supporto alle metodologie allo stato dell’arte di Deep Learning. Tramite Torch7 è stata implementata la vera e propria classificazione di immagini poiché questo framework, grazie anche al lavoro di analisi portato avanti da alcuni miei colleghi in precedenza, è risultato essere molto efficace nel categorizzare oggetti in immagini. Le immagini su cui si sono basati i test sperimentali, appartengono a un dataset creato ad hoc per il sistema di visione 3D con la finalità di sperimentare il sistema per individui ipovedenti e non vedenti; in esso sono presenti alcuni tra i principali ostacoli che un ipovedente può incontrare nella propria quotidianità. In particolare il dataset si compone di potenziali ostacoli relativi a una ipotetica situazione di utilizzo all’aperto. Dopo avere stabilito dunque che Torch7 fosse il supporto da usare per la classificazione, l’attenzione si è concentrata sulla possibilità di sfruttare la Visione Stereo per aumentare l’accuratezza della classificazione stessa. Infatti, le immagini appartenenti al dataset sopra citato sono state acquisite mediante una Stereo Camera con elaborazione su FPGA sviluppata dal gruppo di ricerca presso il quale è stato svolto questo lavoro. Ciò ha permesso di utilizzare informazioni di tipo 3D, quali il livello di depth (profondità) di ogni oggetto appartenente all’immagine, per segmentare, attraverso un algoritmo realizzato in C++, gli oggetti di interesse, escludendo il resto della scena. L’ultima fase del lavoro è stata quella di testare Torch7 sul dataset di immagini, preventivamente segmentate attraverso l’algoritmo di segmentazione appena delineato, al fine di eseguire il riconoscimento della tipologia di ostacolo individuato dal sistema.

Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

The alphavbeta6 integrin is a highly specific immunohistochemical marker for cholangiocarcinoma

Hepatocellular carcinoma (HCC) and cholangiocarcinoma (CC) are common primary hepatic malignancies. Their immunohistological differentiation using specific markers is pivotal for treatment and prognosis. We found alphavbeta6 integrin strongly upregulated in biliary fibrosis, but its expression in primary and secondary liver tumours is unknown. Here, we aimed to evaluate the diagnostic applicability of alphavbeta6 integrin in differentiating primary liver cancers.

Continuous Sunitinib treatment in patients with advanced hepatocellular carcinoma: a Swiss Group for Clinical Cancer Research (SAKK) and Swiss Association for the Study of the Liver (SASL) multicenter phase II trial (SAKK 77/06)

Sunitinib (SU) is a multitargeted tyrosine kinase inhibitor with antitumor and antiangiogenic activity. The objective of this trial was to demonstrate antitumor activity of continuous SU treatment in patients with hepatocellular carcinoma (HCC).

Everolimus augments the effects of sorafenib in a syngeneic orthotopic model of hepatocellular carcinoma

Sorafenib targets the Raf/mitogen-activated protein kinase, VEGF, and platelet-derived growth factor pathways and prolongs survival patients in advanced hepatocellular carcinoma (HCC). Everolimus inhibits the mammalian target of rapamycin, a kinase overactive in HCC. To investigate whether the antitumor effects of these agents are additive, we compared a combined and sequential treatment regimen of everolimus and sorafenib with monotherapy. After hepatic implantation of Morris Hepatoma (MH) cells, rats were randomly allocated to everolimus (5 mg/kg, 2×/week), sorafenib (7.5 mg/kg/d), combined everolimus and sorafenib, sequential sorafenib (2 weeks) then everolimus (3 weeks), or control groups. MRI quantified tumor volumes. Erk1/2, 4E-BP1, and their phosphorylated forms were quantified by immunoblotting. Angiogenesis was assessed in vitro by aortic ring and tube formation assays, and in vivo with Vegf-a mRNA and vascular casts. After 35 days, tumor volumes were reduced by 60%, 85%, and 55%, relative to controls, in everolimus, the combination, and sequential groups, respectively (P < 0.01). Survival was longest in the combination group (P < 0.001). Phosphorylation of 4E-BP1 and Erk1/2 decreased after everolimus and sorafenib, respectively. Angiogenesis decreased after all treatments (P < 0.05), although sorafenib increased Vegf-a mRNA in liver tumors. Vessel sprouting was abundant in control tumors, lower after sorafenib, and absent after the combination. Intussusceptive angiogenic transluminal pillars failed to coalesce after the combination. Combined treatment with everolimus and sorafenib exerts a stronger antitumoral effect on MH tumors than monotherapy. Everolimus retains antitumoral properties when administered sequentially after sorafenib. This supports the clinical use of everolimus in HCC, both in combination with sorafenib or after sorafenib.

The homeobox gene HLXB9 is upregulated in a morphological subset of poorly differentiated hepatocellular carcinoma

The prognostic outcome for hepatocellular carcinoma (HCC) remains poor. Disease progression is accompanied by dedifferentiation of the carcinoma, a process that is not well understood. The aim of this study was to get more insight into the molecular characteristics of dedifferentiated carcinomas using high throughput techniques. Microarray-based global gene expression analysis was performed on five poorly differentiated HCC cell lines compared with non-neoplastic hepatic controls and a set of three cholangiolar carcinoma (CC) cell lines. The gene with the highest upregulation was HLXB9. HLXB9 is a gene of the homeobox genfamily important for the development of the pancreas. RT-PCR confirmed the upregulation of HLXB9 in surgical specimens of carcinoma tissue, suggesting its biological significance. Interestingly, HLXB9 upregulation was primary observed in poorly differentiated HCC with a pseudoglandular pattern compared with a solid pattern HCC or in moderate or well-differentiated HCC. Additional the expression of translated HLXB9, the protein HB9 (NCBI: NP_001158727), was analyzed by western blotting. Expression of HB9 was only detected in the cytoplasm but not in the nuclei of the HCC cells. For validation CC were also investigated. Again, we found an upregulation of HLXB9 in CC cells accompanied by an expression of HB9 in the cytoplasms of these tumor cells, respectively. In conclusion, homeobox HLXB9 is upregulated in poorly differentiated HCC with a pseudoglandular pattern. The translated HB9 protein is found in the cytoplasm of these HCC and CC. We therefore assume HLXB9 as a possible link in the understanding of the development of HCC and CC, respectively.

Heterogeneity of patients with intermediate (BCLC B) Hepatocellular Carcinoma: proposal for a subclassification to facilitate treatment decisions

The intermediate stage of hepatocellular carcinoma (HCC) comprises a highly heterogeneous patient population and therefore poses unique challenges for therapeutic management, different from the early and advanced stages. Patients classified as having intermediate HCC by the Barcelona Clinic Liver Cancer (BCLC) staging system present with varying tumor burden and liver function. Transarterial chemoembolization (TACE) is currently recommended as the standard of care in this setting, but there is considerable variation in the clinical benefit patients derive from this treatment.In April 2012, a panel of experts convened to discuss unresolved issues surrounding the application of current guidelines when managing patients with intermediate HCC. The meeting explored the applicability of a subclassification system for intermediate HCC patients to tailor therapeutic interventions based on the evidence available to date and expert opinion. The present report summarizes the proposal of the expert panel: four substages of intermediate HCC patients, B1 to B4.

Characteristics of patients with chronic hepatitis C who develop hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is the most frequent form of primary liver cancer and chronic infection with hepatitis C virus is one of the main risk factors for HCC. This study analyses the characteristics of the patients with chronic hepatitis C participating in the Swiss Hepatitis C Cohort Study who developed HCC.

Plasma biomarkers as predictors of outcome in patients with advanced hepatocellular carcinoma

Validated biomarkers of prognosis and response to drug have not been identified for patients with hepatocellular carcinoma (HCC). One of the objectives of the phase III, randomized, controlled Sorafenib HCC Assessment Randomized Protocol (SHARP) trial was to explore the ability of plasma biomarkers to predict prognosis and therapeutic efficacy.

Hepatocellular carcinoma, syncytial giant cell: a novel variant in children: a case report

Hepatocellular carcinoma (HCC) is the second most common primary malignant hepatic tumor in children. It often develops in patients with underlying liver disease. We report the clinicopathologic features of an unusual HCC occurring in an infant who presented with features of Cushing's syndrome due to bilateral adrenal hyperplasia. The tumor is characterized by epithelial syncytial giant cells. Giant cell carcinoma of the liver has been previously reported, but the cells were osteoclast-like (ie, mesenchymal type) and not epithelial type as it is in this patient. We propose to use the term HCC, syncytial giant cell type, to denote this apparently novel lesion.

Vascular remodeling and antitumoral effects of mTOR inhibition in a rat model of hepatocellular carcinoma

BACKGROUND/AIMS: Hepatocellular carcinoma (HCC) is amenable to only few treatments. Inhibitors of the kinase mTOR are a new class of immunosuppressors already in use after liver transplantation. Their antiproliferative and antiangiogenic properties suggest that these drugs could be considered to treat HCC. We investigated the antitumoral effects of mTOR inhibition in a HCC model. METHODS: Hepatoma cells were implanted into livers of syngeneic rats. Animals were treated with the mTOR inhibitor sirolimus for 4 weeks. Tumor growth was monitored by MR imaging. Antiangiogenic effects were assessed in vivo by microvessel density and corrosion casts and in vitro by cell proliferation, tube formation and aortic ring assays. RESULTS: Treated rats had significantly longer survival and developed smaller tumors, fewer extrahepatic metastases and less ascites than controls. Sirolimus decreased intratumoral microvessel density resulting in extensive necrosis. Endothelial cell proliferation was inhibited at lower drug concentrations than hepatoma cells. Tube formation and vascular sprouting of aortic rings were significantly impaired by mTOR inhibition. Casts revealed that in tumors treated with sirolimus vascular sprouting was absent, whereas intussusception was observed. CONCLUSIONS: mTOR inhibition significantly reduces HCC growth and improves survival primarily via antiangiogenic effects. Inhibitors of mTOR may have a role in HCC treatment.

Hypoxia increases cytoplasmic expression of NDRG1, but is insufficient for its membrane localization in human hepatocellular carcinoma

NDRG1 is a hypoxia-inducible protein, whose modulated expression is associated with the progression of human cancers. Here, we reveal that NDRG1 is markedly upregulated in the cytoplasm and on the membrane in human hepatocellular carcinoma (HCC). We demonstrate further that hypoxic stress increases the cytoplasmic expression of NDRG1 in vitro, but does not result in its localization on the plasma membrane. However, grown within an HCC-xenograft in vivo, cells express NDRG1 in the cytoplasm and on the plasma membrane. In conclusion, hypoxia is a potent inducer of NDRG1 in HCCs, albeit requiring additional stimuli within the tumour microenvironment for its recruitment to the membrane.

Radiological diagnosis of hepatocellular carcinoma

Treatment of hepatocellular carcinomas (HCC) is often complicated by the fact that early HCCs are mostly asymptomatic and the carcinoma is often discovered at an advanced stage. The aim of diagnostic imaging is to detect HCC at an early stage, when curative options are available. In recent years, there have been many efforts to improve early detection of small HCC. The purpose of this article is to describe the pertinent findings of HCCs in non-invasive, diagnostic imaging, including ultrasound, computed tomography, as well as modern magnetic resonance imaging techniques. Special emphasis is given to the frequently addressed difficulties of differentiation of precancerous lesions and small HCCs. A non-invasive diagnostic approach is considered with a review of the literature.

Losses of chromosome arms 4q, 8p, 13q and gain of 8q are correlated with increasing chromosomal instability in hepatocellular carcinoma

OBJECTIVE: Chromosomal instability is a key feature in hepatocellular carcinoma (HCC). Array comparative genomic hybridization (aCGH) revealed recurring structural aberrations, whereas fluorescence in situ hybridization (FISH) indicated an increasing number of numerical aberrations in dedifferentiating HCC. Therefore, we examined whether there was a correlation between structural and numerical aberrations of chromosomal instability in HCC. METHODS AND RESULTS: 27 HCC (5 well, 10 moderately, 12 lower differentiated) already cytogenetically characterized by aCGH were analyzed. FISH analysis using probes for chromosomes 1, 3, 7, 8 and 17 revealed 1.46-4.24 signals/nucleus, which correlated with the histological grade (well vs. moderately,p < 0.0003; moderately vs. lower, p < 0.004). The number of chromosomes to each other was stable with exceptions only seen for chromosome 8. Loss of 4q and 13q, respectively, were correlated with the number of aberrations detected by aCGH (p < 0.001, p < 0.005; Mann-Whitney test). Loss of 4q and gain of 8q were correlated with an increasing number of numerical aberrations detected by FISH (p < 0.020, p < 0.031). Loss of 8p was correlated with the number of structural imbalances seen in aCGH (p < 0.048), but not with the number of numerical changes seen in FISH. CONCLUSION: We found that losses of 4q, 8p and 13q were closely correlated with an increasing number of aberrations detected by aCGH, whereas a loss of 4q and a gain of 8q were also observed in the context of polyploidization, the cytogenetic correlate of morphological dedifferentiation.