Hepatocellular carcinoma (HCC) is one of the commonest fatal tumors, and it is usually diagnosed at a late stage, when effective treatment is very difficult. Unfortunately early diagnosis of HCC is almost mandatory in terms of patient survival, but is represents a very difficult task. Detailed histological characteristics of small HCC and precursor lesions in histological diagnosis are needed to refine the contribution of histopathology to the management of patients with HCC. In this context a possible approach regards the implementation of knowledge extraction process based on a labeled data set of cases. Our data set is made of 212 cases, including early malignant nodules, benign ones, and some uncertain cases. A set of 11 histopathological and morphological features described each case. We applied statistical and machine learning algorithms both for feature selection and classification of nodules. In particular, we implemented an ensemble of classifiers in a straightforward way. Moreover, this ensemble provides physicians with a rough level of confidence of the proposed diagnoses, taking into account the importance of the interpretability of automated output. Feature selection approach was validated; actually a subset of four histological features performed at least as the whole set of features in the classification of the malignant and benign nodules. Ensemble of learning machines can be a tool to extract new knowledge from a set of data and improve comprehensibility of computerized output.

Knowledge Discovery in Support of Early Diagnosis of Hepatocellular Carcinoma

Ciocchetta, Federica;Dell'Anna, Rossana;Demichelis, Francesca;Sboner, Andrea;
2003

Abstract

Hepatocellular carcinoma (HCC) is one of the commonest fatal tumors, and it is usually diagnosed at a late stage, when effective treatment is very difficult. Unfortunately early diagnosis of HCC is almost mandatory in terms of patient survival, but is represents a very difficult task. Detailed histological characteristics of small HCC and precursor lesions in histological diagnosis are needed to refine the contribution of histopathology to the management of patients with HCC. In this context a possible approach regards the implementation of knowledge extraction process based on a labeled data set of cases. Our data set is made of 212 cases, including early malignant nodules, benign ones, and some uncertain cases. A set of 11 histopathological and morphological features described each case. We applied statistical and machine learning algorithms both for feature selection and classification of nodules. In particular, we implemented an ensemble of classifiers in a straightforward way. Moreover, this ensemble provides physicians with a rough level of confidence of the proposed diagnoses, taking into account the importance of the interpretability of automated output. Feature selection approach was validated; actually a subset of four histological features performed at least as the whole set of features in the classification of the malignant and benign nodules. Ensemble of learning machines can be a tool to extract new knowledge from a set of data and improve comprehensibility of computerized output.
0-7803-7898-9
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11582/4441
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact