Search

In this paper, we investigate the impact of optimal hyper-parameter configuration in relational topic models. The main goal is to validate the hypothesis that single-objective Bayesian Optimization (BO) can discover a hyper-parameter setting that leads a set of relational topic models to simultaneously ensure good prediction capabilities and significant topics from a qualitative perspective. Our research, as a result of a comparative analysis performed on 7 state-of-the-art models, 5 performance measures and 3 datasets, has highlighted three main findings: (1) the majority of relational topic models are not able to offer a good trade-off between classification capabilities and topic interpretability; (2) single-objective optimization of hyper-parameters, targeted on maximizing the F1-Measure, is able to create topics that are also optimal with respect to the Kullback Leibler divergence measure; (3) the Pareto frontiers across several performance metrics reveals that the most promising trade-off between the performance metrics can be obtained by Constrained Relational Topic Models.

Undifferentiated pancreatic carcinoma, also called anaplastic pancreatic carcinoma or giant cells pancreatic carcinoma, is an uncommon and aggressive variety of ductal adenocarcinoma. Three types of undifferentiated carcinoma of the pancreas are known: Osteclastic-like giant cells tumor, pleomorphic giant cells tumor, and mixed giant cells tumor. Osteoclast-like giant cells tumor is characterized by the presence of large histocytic elements with many small nuclei showing no atypia. Pleomorphic giant cells tumor is characterized by the presence of bizarre mono- or multinucleated giant cells with atypical mitoses. The mixed giant cells tumor shows the simultaneous presence of both histological histotypes. The histogenesis of giant cells is controversial. Currently osteoclast-like giant cells are considered reactive elements deriving from a histiocytic lineage. Pleomorphic giant cells tumor originates from ductal epithelium with subsequent sarcomatous transformation. The various hystotypes show different behavior in terms of survival. Pleomorphic giant cells tumor is characterized by a worse prognosis than ductal adenocarcinoma while osteoclast-like giant cells tumor can be associated with longer survival time. The mixed giant cells tumor presents an intermediate prognosis. In conclusion, in case of undifferentiated pancreatic carcinoma, an accurate histopathological diagnosis can predict different behavioral pathways in terms of tumor progression and prognostic profile.

Il condiloma acuminato gigante (GCA) è una rara lesione simile a cavolfiore, nota anche come tumore di Buschke-Lewenstein (BLT). Sebbene sia caratterizzato da caratteristiche istologiche benigne, il comportamento locale del GCA può essre estremamente aggressivo, mostrando una progressiva infiltrazione delle strutture circostanti che portano alla distruzione del tessuto mediante compressione. Poiché la correlazione tra lo sviluppo di HPV e GCA è aumentata, la maggior parte degli Autori è giunta alla conclusione che l’HPV non solo può causare condilomatosi, ma, associato a particolari fattori di rischio, può portare a condizioni molto più gravi come BLT. Poiché il trattamento della GCA non è ancora standardizzato, è necessaria un’analisi preoperatoria molto accurata delle lesioni per pianificare l’approccio terapeutico più adatto. Sulla base delle attuali conoscenze, la valutazione macroscopica dell’invasione locale del tumore e l’estesa resezione radicale sembrano essere l’unico approccio terapeutico valido. In attesa di nuove tecniche e nuovi trattamenti di risparmio tissutale, al momento, la GCA dovrebbe essere trattata con un’escissione radicale senza una ricostruzione tissutale immediata; complicazioni a lungo termine, come la stenosi, possono essere prevenute con un’adeguata guarigione delle ferite e con un programma di follow-up particolarmente intenso e a lungo termine. Giant condyloma acuminatum (GCA) is a rare cauliflower-like lesion, also known as Buschke-L ewenstein tumor (BLT). Although characterized by benign histological features, the local behavior of GCA is extremely aggressive, showing progressive infiltration of the surrounding structures leading to tissue destruction by compression. As the correlation between HPV and GCA development grew stronger, the majority of the Authors came to the conclusion that HPV can not only cause CA but, associated with particular risk factors, it can lead to much more serious conditions such as BLT. Since the treatment of GCA is still not yet standardized, a very accurate pre-operative analysis of the lesions is required to plan the most suitable treatment approach. Based on current knowledge, macroscopic evaluation of local tumor invasion and extensive radical resection appear to be the only valid therapeutic approach, due to its association with longterm survival and minimal recurrence. Looking forward for new techniques and new tissue sparing treatments, at the moment, GCA can be safely treated with radical excision without immediate tissue reconstruction; long-term complications, such as stricture and stenosis, can be prevented by adequate wound healing and by a particularly intense and long- term follow-up program. KEY WORDS: Anus, Giant condyloma acuminatum.

La stenosi anale è un restringimento fibroso del canale anale. Si sviluppa, nella maggior parte dei casi, in seguito a procedure chirurgiche proctologiche con ampia escissione anodermica o in presenza di infiammazione anale cronica in pazienti con malattia di Crohn. Per quanto rara, questa condizione è molto debilitante per i pazienti. I sintomi includono stitichezza, dolore e defecazione con anguinamento e una riduzione del calibro delle feci. La diagnosi è essenzialmente clinica. La prevenzione della stenosi post-chirurgica si basa su una scrupolosa tecnica chirurgica e su un follow-up ampio e attentamente pianificato. Il trattamento si basa su un approccio inizialmente conservativo con la regolarizzazione del transito delle feci attraverso l’idratazione, fibre alimentari e lassativi formanti massa. Il ruolo della dilatazione meccanica nel trattamento della stenosi anale è ancora oggetto di discussione. Per i casi gravi che non rispondono al trattamento conservativo si deve ricorrere alla chirurgia. L’approccio chirurgico è adattato alla gravità e all’estensione fino ad arrivare all’anoplastica con flap mucoso. Anal stenosis is a fibrous narrowing of the anal channel. It develops, in most cases, in the aftermath of proctologic surgical procedures with extensive anoderm excision or in the presence of chronic anal inflammation in patients with Crohn Disease. However rare, this condition is thoroughly debilitating for the patients. Symptoms include constipation, pain and bleeding with defecation and a reduction of the caliber of stools. Diagnosis is essentially clinical. Prevention of postsurgical stenosis is based on a scrupulous surgical technique and on an extensive and carefully- planned follow up. Treatment is based on an initially conservative approach with regularization of stool transit through hydration, dietary fibers and bulk-forming laxatives. The role of mechanical dilatation in the treatment of AS is still debated. For severe cases and cases that are not responsive to conservative treatment we must resort to surgery. Surgical approach is tailored on the severity, position and extension. The most severe cases require anoplasty procedures with mucosal or anal flaps. KEY WORDS: Anal stenosis, Anal stricture,Proctology.

Relational topic models (RTM) have been widely used to discover hidden topics in a collection of networked documents. In this paper, we introduce the class of Constrained Relational Topic Models (CRTM), a semi-supervised extension of RTM that, apart from modeling the structure of the document network, explicitly models some available domain knowledge. We propose two instances of CRTM that incorporate prior knowledge in the form of document constraints. The models smooth the probability distribution of topics such that two constrained documents can either share the same topics or denote distinct themes. Experimental results on benchmark relational datasets show significant performances of CRTM on a semi-supervised document classification task.

La conoscenza testuale è uno dei pilastri principali della nostra società. Infatti, la conoscenza umana è spesso trasmessa attraverso le parole. Dall'invenzione della scrittura, gli uomini hanno raccontato e descritto la loro esistenza con parole su pezzi di carta. Questa quantità di conoscenza si accumula fino a ciò che l'intera civiltà ha raccolto in più di 5000 anni. Gli storici e gli scienziati sociali e politici cercano modi per capire meglio questa vasta quantità di conoscenza collettiva che non può essere esplorata manualmente. A questo scopo, i ricercatori di machine learning, statistica e linguistica computazionale hanno sviluppato i topic model, una suite di algoritmi che mirano ad annotare grandi archivi di documenti con informazioni tematiche. La popolarità di questi modelli è dovuta al fatto che sono non supervisionati e che sono interpretabili. I topic model analizzano e riassumono i temi principali, anche detti topic, di grandi collezioni di documenti, presentando le informazioni in una forma compatta e comprensibile. La maggior parte dei topic model si concentrano solo sulle parole codificate nei documenti. Tuttavia, informazioni aggiuntive possono essere introdotte nei topic model per migliorare le loro prestazioni. Infatti, in molti casi del mondo reale, raramente abbiamo solo i semplici testi da analizzare. Invece, abbiamo informazioni aggiuntive o metadati relativi ai documenti, ad esempio, l'autore del documento, la data, collegamenti ipertestuali ad altri documenti, un insieme di hashtag, menzioni o etichette. Possiamo usare queste informazioni precedenti per aiutare un topic model a scoprire topic di qualità. Per esempio, sapere che un documento cita un altro documento aumenta la nostra fiducia che i documenti parlano degli stessi argomenti.Inoltre, i topic model spesso ignorano l'ordine delle parole e le informazioni contestuali, rendendo difficile dedurre gli topic coerenti e significativi. Un altro problema nel campo è legato agli iperparametri usati per durante il processo di learning dei topic model. Gli iperparametri sono spesso fissati e pochi ricercatori hanno cercato di studiare il loro impatto sui risultati. In questa tesi, ci proponiamo di affrontare i problemi menzionati. Introduciamo nuove famiglie di topic model per ottenere prestazioni migliori. Esploriamo anche le questioni relative all'ottimizzazione degli iperparametri progettando e sviluppando un nuovo strumento per fornire ai ricercatori delle linee guida migliori su come usare un topic model. Textual knowledge is one of the main pillars of our society. Indeed, human knowledge is often passed along using words. Since the invention of writing, humans have narrated and described their existence with words over pieces of papers. This amount of knowledge builds up to what the entire civilization has collected over more than 5'000 years. Historians and social and political scientists look for ways to understand better this vast amount of collective knowledge that cannot be manually explored. To this end, researchers from machine learning, statistics and computational linguistic have developed topic models, a suite of algorithms that aim to annotate large archives of documents with thematic information. The popularity of these models is due to the fact that they are unsupervised and that they are interpretable. Topic models analyze and summarize the main themes, or topics, of large collections of documents, presenting the information in a compact and understandable form. Most topic models focus only on the words encoded in the documents. However, additional information can be introduced into topic models to improve their performance. In fact, in many real-world cases, we seldom have only the mere texts to analyze. Instead, we have additional information or metadata related to the documents, e.g., the document's author, the date, hyperlinks to other documents, a set of hashtags, mentions or labels. We can use this prior information to help a topic model discover better topics. For example, knowing that a document cites another document increases our confidence that the documents talk about the same topics. Also, topic models often ignore word order and contextual information, making it difficult to infer high-quality topics. Another problem in the field is related to the hyperparameters used to train the topics models. The hyperparameters are often fixed in experimental settings and few researchers have tried to study their impact on the results. In this thesis, we aim to tackle the mentioned problems. We introduce novel families of topic models to obtain better performance. We also explore the issues related to hyperparameter optimization by designing and developing a novel tool to supply researchers with better guidelines on how to train a topic model.

Many data sets (e.g., reviews, forums, news, etc.) exist parallelly in multiple languages. They all cover the same content, but the linguistic differences make it impossible to use traditional, bag-of-word-based topic models. Models have to be either single-language or suffer from a huge, but extremely sparse vocabulary. Both issues can be addressed by transfer learning. In this paper, we introduce a zero-shot cross-lingual topic model. Our model learns topics on one language (here, English), and predicts them for unseen documents in different languages (here, Italian, French, German, and Portuguese). We evaluate the quality of the topic predictions for the same document in different languages. Our results show that the transferred topics are coherent and stable across languages, which suggests exciting future research directions., Comment: Updated version. Published as a conference paper at EACL2021

Neural Topic Models are recent neural models that aim at extracting the main themes from a collection of documents. The comparison of these models is usually limited because the hyperparameters are held fixed. In this paper, we present an empirical analysis and comparison of Neural Topic Models by finding the optimal hyperparameters of each model for four different performance measures adopting a single-objective Bayesian optimization. This allows us to determine the robustness of a topic model for several evaluation metrics. We also empirically show the effect of the length of the documents on different optimized metrics and discover which evaluation metrics are in conflict or agreement with each other.

Topic models aim at discovering a set of hidden themes in a text corpus. A user might be interested in identifying the most similar topics of a given theme of interest. To accomplish this task, several similarity and distance metrics can be adopted. In this paper, we provide a comparison of the state-of-the-art topic similarity measures and propose novel metrics based on word embeddings. The proposed measures can overcome some limitations of the existing approaches, highlighting good capabilities in terms of several topic performance measures on benchmark datasets.

OCTIS is an open-source framework for training, evaluating and comparing Topic Models. This tool uses single-objective Bayesian Optimization (BO) to optimize the hyper-parameters of the models and thus guarantee a fairer comparison. Yet, a single-objective approach disregards that a user may want to simultaneously optimize multiple objectives. We therefore propose OCTIS 2.0: the extension of OCTIS that addresses the problem of estimating the optimal hyper-parameter configurations for a topic model using multi-objective BO. Moreover, we also release and integrate two pre-processed Italian datasets, which can be easily used as benchmarks for the Italian language. OCTIS è un framework open-source per il training, la valutazione e la comparazione di Topic Models. Questo strumento utilizza l’ottimizzazione Bayesiana (BO) a singolo obiettivo per ottimizzare gli iperparametri dei modelli e quindi garantire una comparazione più equa. Tuttavia, questo approccio ignora che un utente potrebbe voler ottimizzare pi‘u di un obiettivo. Proponiamo perciò OCTIS 2.0: l’estensione di OCTIS che affronta il problema della stima delle configurazioni ottimali degli iperparametri di un topic model usando la BO multi-obiettivo. In aggiunta, rilasciamo e integriamo anche due nuovi dataset in italiano pre-processati, che possono essere facilmente utilizzati come benchmark per la lingua italiana.

In this paper, we present OCTIS, a framework for training, analyzing, and comparing Topic Models, whose optimal hyper-parameters are estimated using a Bayesian Optimization approach. The proposed solution integrates several state-of-the-art topic models and evaluation metrics. These metrics can be targeted as objective by the underlying optimization procedure to determine the best hyper-parameter configuration. OCTIS allows researchers and practitioners to have a fair comparison between topic models of interest, using several benchmark datasets and well-known evaluation metrics, to integrate novel algorithms, and to have an interactive visualization of the results for understanding the behavior of each model. The code is available at the following link: https://github.com/MIND-Lab/OCTIS.

SCOPO: Obiettivo dello studio descrivere l’incidenza, le caratteristiche radiologiche e anatomo-patologiche degli incidentalomi pancreatici. Nell’articolo, Inoltre, sono discusse le indicazioni chirurgiche di queste lesioni in base alla loro natura e localizzazione. PREMESSE: Le lesioni pancreatiche incidentali sono sempre pi frequentemente diagnosticate grazie alla larga diffusione di tecniche radiologiche di elevate qualit . Gli incidentalomi pancreatici possono essere cistici o solidi, benigni o maligni, coprendo un ampio spettro di variet istologiche. Le lesioni cistiche sono pi frequentemente di natura benigna o pre-maligna. L’asportazione chirurgica deve essere riservata solo in caso di variazioni sospette in senso degenerativo al follow-up oppure per le masse cistiche di larghe dimensioni (>3 cm). Tra gli incidentalomi pancreatici solidi l’adenocarcinoma duttale la neoplasia pi frequentemente diagnosticata, seguita per frequenza dei tumori neuroendocrini. Il trattamento chirurgico dipende dalla localizzazione dell’incidentaloma: la duodenocefalopancreasectomia dovrebbe essere eseguita per le lesioni a carico della testa, mentre la pancreasectomia distale indicata per quelle del corpo e della coda. La chirurgia del pancreas delicata e tuttora gravata da serie complicazioni. Entrambi gli interventi chirurgici possono eseguiti con tecnica mini-invasiva, che caratterizzata da un minor tasso di complicazioni, ma al momento attuale, non stata dimostrata una miglior sopravvivenza nei pazienti oncologici CONCLUSIONI: Gli incidentalomi pancreatici sono di riscontro sempre pi frequente nella pratica clinica, ma quando e come intervenire chirurgicamente ancora materia di dibattito. Attualmente mancano precisi criteri circa la strategia di trattamento e sono quindi necessarie linee-guida per definire il miglior approccio di queste lesioni del pancreas. AIM: The aim of this study is to describe the incidence, imaging characteristics and pathological features of pancreatic incidentalomas. Moreover, surgical indications are discussed according to the nature and location of the neoplasms. BACKGROUND: Pancreatic incidental lesions are more commonly diagnosed, due to the widespread of high quality cross sectional imaging. These lesions can be cystic or solid, benign, pre-malignant or already malignant and they cover a wide spectrum of histological diagnosis. Cystic lesions are more commonly benign or at least pre-malignant. Surgery should be reserved in case of unexpected changes in aspect during follow-up or for large cysts (>3 cm). Among solid pancreatic incidentalomas, ductal adenocarcinoma is the most common diagnosis, followed by neuroendocrine tumors. Surgical treatment of pancreatic incidentaloma depends on the location of the tumor: a Whipple's procedure should be performed for neoplasms of the head, while distal pancreatectomy is indicated for body and tail lesions. Pancreatic surgery is still delicate and burdened by serious complications. Both procedures can be performed with minimally-invasive technique which is connected to lower complications rate but, at present, they have shown no advantages in terms of mortality and oncologic outcomes. CONCLUSIONS: Pancreatic incidentalomas are becoming more and more common but when and how to operate them is still subject of debate. Precise criteria about treatment strategy are still lacking and definite guidelines are needed to clarify the best approach.

Topic models have been widely used to discover hidden topics in a collection of documents. In this paper, we propose to investigate the role of two different types of relational information, i.e. document relationships and concept relationships. While exploiting the document network significantly improves topic coherence, the introduction of concepts and their relationships does not influence the results both quantitatively and qualitatively.

Background The burden of disease caused by Streptococcus agalactiae has increased significantly among older adults in the last decades. Group B streptococcus infection can be associated with invasive disease and severe clinical syndromes, such as meningitis and endocarditis. Case presentation We present the case of a 56-year-old man who developed multiple mycotic aneurysms of the right hepatic artery and massive splenic infarction as rare complications of Streptococcus agalactiae infective endocarditis. The patient underwent urgent right hepatic artery ligation and splenectomy. The postoperative course was complicated by an episode of hemobilia due to the rupture of a partially thrombosed mycotic aneurysm into the biliary tree. Thus, selective radiological embolization of the left hepatic artery branches was necessary. Conclusion To our knowledge, this is the first case reported of infected aneurysms of visceral arteries caused by Group B streptococcus infection. Clinical and laboratory findings were non-specific, while imaging features with computed tomography scan and angiography were highly suggestive. In our case, early recognition, culture-specific intravenous antibiotics and urgent surgical treatment combined with interventional radiology played a decisive role in the final result.

Since the 1970s, in the USA, we witnessed a progressive increase of one-day surgical procedures. This attitude soon gained ground in Europe as well. In proctology, this kind of clinical approach has always been limited by the acute sensitivity of the anal- perineal area and by difficulties in attaining a complete sphincter relaxation with local anesthesia. Posterior perineal block seems to be associated with both a good pain control and an effective sphincter relaxation.Between January 2017 and January 2018, we enrolled in our study 33 patients suffering from hemorrhoidal disease. They were all subjected to posterior perineal block. We measured anal resting pressure and squeeze pressure before and after anesthesia. Measurements where taken 5 minutes before and 15 minutes after the administration of local.We registered an average decrease of 39,2% of resting pressure and of 45,4% of squeeze pressure.We may state that perineal posterior block, while reducing striated muscle contractile activity, also causes a relevant reduction of anal basal tone. During surgical procedures done under regional anesthesia, we experienced a good sphincter relaxation, which was comparable, if not equal, to that induced by general anesthesia. In fact, 10 to 15 minutes after performing the block you could observe the elevation of the inferior margin of the exterior sphincter and the concomitant descent of the inferior margin of the internal sphincter (coaxial dislocation).Anorectal manometry, Anesthesia, Local-regional, Perineal block.A partire dagli anni 70 negli Stati Uniti si è assistito ad un progressivo aumento delle procedure chirurgiche in day surgery Questo processo si è diffuso anche nella maggior parte delle strutture sanitarie in Europa. In ambito proctologico le indicazioni sono sempre state limitate dalla particolare sensibilità della zona ano-perineale e da un difficile completo rilassamento dell’apparato sfinteriale dopo anestesia locale. Il blocco perineale posteriore sembra essere associato oltre che ad una buona analgesia ad un efficace rilassamento sfinterico.Da gennaio 2017 a gennaio 2018 sono stati arruolati 33 pazienti affetti da patologia emorroidaria. In tutti i pazienti è stato eseguito un blocco perineale posteriore. Sono state misurate la pressione basale del canale anale e la contrazione massima volontaria prima e dopo l’anestesia. Le due rilevazioni manometriche sono avvenute in posizione litotomica 5 minuti prima e 15 minuti dopo l’esecuzione dell’anestesia.La diminuzione media del tono basale dopo blocco perineale posteriore è stata pari al 39,2% e la riduzione media della contrazione massima è stata del 45,4%.Possiamo affermare che il blocco perineale posteriore determina oltre alla riduzione dell’attività contattile della muscolatura striata, anche un rilevante decremento del tono anale di base. Durante interventi proctologici eseguiti in anestesia locoregionale con il blocco perineale posteriore si è riscontrato notevole rilasciamento della parete sfinteriale dell’ano paragonabile, anche se non uguale, a quello che si verifica durante un’anestesia generale. Infatti dopo circa 10-15 minuti nella quasi totalità dei casi si osserva sia la risalita del margine inferiore dello sfintere esterno che la concomitante discesa del margine inferiore dello sfintere interno (dislocazione coassiale).

Background Pancreatic duct disruption is a challenging condition leading to pancreatic juice leakage and consequently to pancreatic fluid collections. The manifestations of pancreatic main duct leak include pseudocysts, walled-off necrosis, pancreatic fistulas, ascites, pleural and pericardial effusions. Pseudocyst formation is the most frequent outcome of a pancreatic duct leak. Case presentation We describe a case of a 64-year old man with large multiple pancreatic cysts discovered for progressive jaundice and significant weight loss in the absence of a previous episode of acute pancreatitis. Computed tomography scan showed lesion with thick enhancing walls. The main cyst dislocated the stomach and the duodenum inducing intra and extrahepatic bile ducts enlargement. Magnetic resonance cholangiopancreatography revealed a communication between the main pancreatic duct and the cystic lesions due to Wirsung duct rupture. Endoscopic ultrasound guided fine needle aspiration cytology did not show neoplastic cells and cyst fluid analysis revealed high amylase concentration. Preoperative exams were suggestive but not conclusive for a benign lesion. Laparotomy was necessary to confirm the presence of large communicating pseudocysts whose drainage was performed by cystogastrostomy. Histology confirmed the inflammatory nature of the cyst wall. Subsequently, the patient had progressive jaundice resolution. Conclusion Pancreatic cystic masses include several pathological entities, ranging from benign to malignant lesions. Rarely pseudocysts present as complex cystic pancreatic lesions with biliary compression in absence of history of acute pancreatitis. We describe the rare case of multiple pancreatic pseudocysts due to Wirsung duct rupture in absence of previous trauma or acute pancreatitis. Magnetic resonance showed the presence of communication with the main pancreatic duct and endoscopic ultrasound fine needle aspiration suggested the benign nature of the lesion.

Chylothorax is a severe condition resulting from the accumulation of chyle into the pleural space. We report the treatment of postoperative chylothorax after resection of mediastinal ganglioneuroma in a 17-year-old boy. Since conservative measures were not effective, we performed direct ligation of lymphatic vessels and pleurodesis. At subsequent surgical re-exploration for persisting chylothorax, accurate inspection of pleural cavity revealed residual chyle leakage. Fibrin sealant patches (TachoSil®) were placed over the source of leak with complete resolution of chylous effusion. To our knowledge, this is the first report of postoperative chylothorax successfully treated by the use of a fibrin sealant patch.