Information

Abstract

Breast cancer is the second most common cancer death among women in developed countries. In less developed countries it has a mortality rate of about 25% rendering it the most common cancer death. It has been demonstrated that an early breast cancer diagnosis significantly reduces the mortality. In addition to mammography and breast ultrasound, Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) is the modality with the highest sensitivity for breast cancer detection. However, systems for automatic lesion analysis are scarce. This thesis proposes a method for lesion evaluation without the necessity of tumor segmentation. The observer has to define a Region Of Interest (ROI) covering the lesion in question and the proposed system performs an automated lesion inspection by computing its Fourier transform. Using the Fourier transformed volume we compute the inertia tensor of its magnitude. Based on the gathered information, the Göttinger score, which is a common breast cancer analysis scheme, is computed and the features are presented in newly create plots. These plots are evaluated with a survey where radiologists participated. The Göttinger score assigns a numeric value for the following features: shape, boundary, Internal Enhancement Characteristics (IEC), Initial Signal Increase (ISI) and Post Initial Signal (PIS). We tested our method on 22 breast tumors (14 malignant and 8 benign ones). Subsequently, we compared our results to the classification of an experienced radiologist. The automatic boundary classification has an accuracy of 0.818, the shape 0.773 and the IEC 0.886 compared to the radiologist’s results. An evaluation of the accuracy of the benign vs. malignant classification shows that the method has an accuracy of 0.682 for all the Göttinger score features and 0.772 using only the shape, boundary and IEC. The evaluation of the plot shows that radiologist like the visual representation of the Göttinger score for single lesions, they, however, refuse the plots where multiple lesions are presented in one visual representation.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@mastersthesis{Hirsch_Christian_2015_ABL,
  title =      "Automatic Breast Lesion Examination of DCE-MRI Data Based on
               Fourier Analysis",
  author =     "Christian Hirsch",
  year =       "2015",
  abstract =   "Breast cancer is the second most common cancer death among
               women in developed countries. In less developed countries it
               has a mortality rate of about 25% rendering it the most
               common cancer death. It has been demonstrated that an early
               breast cancer diagnosis significantly reduces the mortality.
               In addition to mammography and breast ultrasound, Dynamic
               Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) is
               the modality with the highest sensitivity for breast cancer
               detection. However, systems for automatic lesion analysis
               are scarce. This thesis proposes a method for lesion
               evaluation without the necessity of tumor segmentation. The
               observer has to define a Region Of Interest (ROI) covering
               the lesion in question and the proposed system performs an
               automated lesion inspection by computing its Fourier
               transform. Using the Fourier transformed volume we compute
               the inertia tensor of its magnitude. Based on the gathered
               information, the G\"{o}ttinger score, which is a common
               breast cancer analysis scheme, is computed and the features
               are presented in newly create plots. These plots are
               evaluated with a survey where radiologists participated. The
               G\"{o}ttinger score assigns a numeric value for the
               following features: shape, boundary, Internal Enhancement
               Characteristics (IEC), Initial Signal Increase (ISI) and
               Post Initial Signal (PIS). We tested our method on 22 breast
               tumors (14 malignant and 8 benign ones). Subsequently, we
               compared our results to the classification of an experienced
               radiologist. The automatic boundary classification has an
               accuracy of 0.818, the shape 0.773 and the IEC 0.886
               compared to the radiologist’s results. An evaluation of
               the accuracy of the benign vs. malignant classification
               shows that the method has an accuracy of 0.682 for all the
               G\"{o}ttinger score features and 0.772 using only the shape,
               boundary and IEC. The evaluation of the plot shows that
               radiologist like the visual representation of the
               G\"{o}ttinger score for single lesions, they, however,
               refuse the plots where multiple lesions are presented in one
               visual representation.",
  month =      sep,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2015/Hirsch_Christian_2015_ABL/",
}