% 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”.

@ARTICLE{Ellmann:157752,
      author       = {S. Ellmann and L. Seyler and C. Gillmann$^*$ and V. Popp
                      and C. Treutlein and A. Bozec and M. Uder and T. Bäuerle},
      title        = {{M}achine {L}earning {A}lgorithms for {E}arly {D}etection
                      of {B}one {M}etastases in an {E}xperimental {R}at {M}odel.},
      journal      = {JoVE journal},
      volume       = {162},
      issn         = {1940-087X},
      address      = {Cambridge, MA},
      publisher    = {JoVE},
      reportid     = {DKFZ-2020-01789},
      pages        = {61235},
      year         = {2020},
      abstract     = {Machine learning (ML) algorithms permit the integration of
                      different features into a model to perform classification or
                      regression tasks with an accuracy exceeding its
                      constituents. This protocol describes the development of an
                      ML algorithm to predict the growth of breast cancer bone
                      macrometastases in a rat model before any abnormalities are
                      observable with standard imaging methods. Such an algorithm
                      can facilitate the detection of early metastatic disease
                      (i.e., micrometastasis) that is regularly missed during
                      staging examinations. The applied metastasis model is
                      site-specific, meaning that the rats develop metastases
                      exclusively in their right hind leg. The model's tumor-take
                      rate is $60\%-80\%,$ with macrometastases becoming visible
                      in magnetic resonance imaging (MRI) and positron emission
                      tomography/computed tomography (PET/CT) in a subset of
                      animals 30 days after induction, whereas a second subset of
                      animals exhibit no tumor growth. Starting from image
                      examinations acquired at an earlier time point, this
                      protocol describes the extraction of features that indicate
                      tissue vascularization detected by MRI, glucose metabolism
                      by PET/CT, and the subsequent determination of the most
                      relevant features for the prediction of macrometastatic
                      disease. These features are then fed into a model-averaged
                      neural network (avNNet) to classify the animals into one of
                      two groups: one that will develop metastases and the other
                      that will not develop any tumors. The protocol also
                      describes the calculation of standard diagnostic parameters,
                      such as overall accuracy, sensitivity, specificity,
                      negative/positive predictive values, likelihood ratios, and
                      the development of a receiver operating characteristic. An
                      advantage of the proposed protocol is its flexibility, as it
                      can be easily adapted to train a plethora of different ML
                      algorithms with adjustable combinations of an unlimited
                      number of features. Moreover, it can be used to analyze
                      different problems in oncology, infection, and
                      inflammation.},
      cin          = {E040},
      ddc          = {570},
      cid          = {I:(DE-He78)E040-20160331},
      pnm          = {315 - Imaging and radiooncology (POF3-315)},
      pid          = {G:(DE-HGF)POF3-315},
      typ          = {PUB:(DE-HGF)16},
      pubmed       = {pmid:32865533},
      doi          = {10.3791/61235},
      url          = {https://inrepo02.dkfz.de/record/157752},
}