h1

% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@PHDTHESIS{Schneider:794608,
      author       = {Schneider, Hannah Sophia},
      othercontributors = {Schrading, Simone and Najjari, Laila},
      title        = {{R}adiomics in der {M}amma {MRT}: {K}lassifikation
                      kontrastmittelaufnehmender {L}äsionen nach ihrer
                      {D}ignität mittels zweier klassischer {M}achine {L}earning
                      {A}lgorithmen},
      school       = {Rheinisch-Westfälische Technische Hochschule Aachen},
      type         = {Dissertation},
      address      = {Aachen},
      reportid     = {RWTH-2020-07774},
      pages        = {1 Online-Ressource (IV, 66 Seiten) : Illustrationen,
                      Diagramme},
      year         = {2020},
      note         = {Veröffentlicht auf dem Publikationsserver der RWTH Aachen
                      University; Dissertation, Rheinisch-Westfälische Technische
                      Hochschule Aachen, 2020},
      abstract     = {In this doctoral thesis, the application of radiomic
                      algorithms in Mamma MRI for classification of
                      contrast-enhancing lesions is depicted. 1294 lesions out of
                      447 breast MRI examinations were manually segmented and 562
                      image features were extracted in total. These included
                      shape, statistical and texture features from routinely
                      acquired MRI sequences. Either histopathological analysis or
                      a 2-year follow-up were used as target values, and for
                      comparison with a radiological expert, lesions were grouped
                      into BI-RADS® categories according to their prospective
                      diagnostic reports. Correlation of the extracted features
                      and the corresponding target values was performed by
                      training and validation of two conventional radiomic
                      algorithms - L1-Regularization and Principal Component
                      Analysis. Though classification results were clinically
                      acceptable (sensitivity $72\%$ and $68\%,$ specificity
                      $76\%$ and $75\%,$ AUC 0,81 and 0,78 respectively), those by
                      the radiologist experts were still superior (sensitivity
                      $100\%,$ specificity $86\%,$ AUC 0,98). A second evaluation
                      on a half-size data set resulted in no deterioration of
                      classificator performance, leading to the conclusion, that
                      an increase in sample size would lead to no improvement by
                      means of a learning system. Considering the selected feature
                      (the so-called radiomic signature), especially features
                      describing shape or intratumoral heterogeneity of contrast
                      enhancement were significant for differentiation of benign
                      vs. malignant lesions. Clinical applications of radiomic
                      algorithms as computer-assisted diagnostic systems might
                      improve diagnostic performance of breast MRI, leading to an
                      increased number of examinations and even an implementation
                      in screening programs. Subsequently, an earlier and better
                      detection of conventionally occult tumors would be possible.
                      Moreover, correlation of image features with tumor-specific
                      parameters, i.e. immunohistochemistry or molecular genetics
                      like the proliferation marker Ki-67 or the OnkotypeDX®
                      Score, might make a future use of image biomarkers as
                      complementary biomarker in precision medicine, for an
                      improved planning of biopsies by depiction of intratumoral
                      heterogeneity or even as virtual biopsies possible. In
                      addition, an improved follow-up of neoadjuvant chemotherapy
                      or adjuvant radiotherapy is conceivable. Until radiomic
                      algorithms can be of clinical use, further development is
                      necessary with promising first results of studies using deep
                      or transfer learning algorithms.},
      cin          = {532010-2},
      ddc          = {610},
      cid          = {$I:(DE-82)532010-2_20140620$},
      typ          = {PUB:(DE-HGF)11},
      doi          = {10.18154/RWTH-2020-07774},
      url          = {https://publications.rwth-aachen.de/record/794608},
}