IMPROVING HEPATIC TUMORS CLASSIFICATION IN MULTIPHASE COMPUTED TOMOGRAPHY USING COMBINED REGIONAL AND SPATIO-TEMPORAL CLASSIFICATION APPROACHES / Shaimaa Ali Shaheen Omer

Supervisor: Sahar A. Fawzi

Thesis (M.A.)—Nile University, Egypt, 2021 .

"Includes bibliographical references"

Contents:
List of Tables .............................................................................................................. vi
List of Figures ............................................................................................................ vii
List of Equations ....................................................................................................... xiii
Abstract ..................................................................................................................... xiv
Introduction ................................................................................................................... 1
Methodology ............................................................................................................... 28
Preprocessing ...................................................................................................29
The Model ........................................................................................................36
Feature Extraction........................................................................................................39
Spatial Approach ..............................................................................................39
Spatio-Temporal Approach ..............................................................................46
Regional Information .......................................................................................53
Results ......................................................................................................................... 61
Discussion and Future Work ....................................................................................... 86
References ...........................

Abstract:
Liver cancer is the most frequently diagnosed cancer in Egypt and North Africa. Early diagnosis provides a better chance for a curative treatment.
Dynamic contrast-enhanced computed tomography (DCE-CT) is the most commonly used technique to differentiate liver tissues. Since, it is available everywhere, accommodates less cost, and takes short scanning time.
In this thesis, DCE-CT images are used to investigate the impact of regional-bases extraction of the spatio-temporal features on the liver lesions classification performance. Texture, density, and temporal feature set, and their combination along spatially partitioned ROI are investigated to characterize five hepatic pathologies from DCE-CT images. Decision tree ensembles (random forest) are employed in both feature selection and classification with ten folds cross-validation to classify a total of 180 ROI. Dataset includes normal tissues, cyst, hemangioma, metastatic and hepatocellular carcinoma (36 ROI for each liver pathology).
The results suggest that normal liver tissues could easily be recognized from just the density features (accuracy = 99%), whereas texture features have the best performance in classifying HCC (accuracy = 96%). Combining all feature sets could overcome individual performance variations among them and attain consistent better results for all tumor types.
Moreover, regional information improves the classification performance in all the classes' results. An overall accuracy of 96% is achieved for the five liver lesions classification.

Text in English, abstracts in English and Arabic