UPDATE  = { 2014-04-07 },
  author =       {Dolej{\v s}{\'\i}, Martin},
  supervisor =   {Kybic, Jan},
  title =        {Detection of Pulmonary Nodules from {CT} Scans},
  school =       {Center for Machine Perception, K13133 FEE Czech Technical
  address =      {Prague, Czech Republic},
  year =         {2014},
  month =        {},
  day =          {},
  type =         {{PhD Thesis CTU--CMP--2014--02}},
  issn =         {1213-2365},
  pages =        {61},
  figures =      {22},
  authorship =   {100},
  psurl =        {[Dolejsi-TR-2014-02.pdf]},
  project =      {SGS10/184/OHK3/2T/13, SGS12/190/OHK3/3T/13,GACR 102/07/1317,
1ET101050403, MSM6840770012},
  annote =       {We present an algorithm to detect small-size (from
                  2~mm to around 10~mm) pulmonary nodules (high
                  intensity objects located in lung parenchyma or
                  attached to the lung wall) from helical CT scans.
                  Lung nodules may indicate a~lung cancer and their
                  detection in the early stage improves the patients
                  survival rate. The large amount of data acquired per
                  CT examination makes the manual interpretation by
                  a~human radiologist time consuming. This leads to
                  a~risk of oversights (false negatives, FNs). The
                  presented method is designed to provide assistance
                  to the radiologist to help reduce the number of
                  FNs. We find the nodule candidates as local maxima
                  in the image filtered by two independent Laplacian
                  of Gaussian filters (LoG) with different variances
                  (11.5 and 200~mm). The candidates are limited to the
                  lung parenchyma by lung segmentation. We segment
                  every candidate detected in the first stage, and
                  compute a feature vector. Finally to determine
                  whether a candidate object is a nodule we classify
                  the candidates represented by their feature vectors
                  by Asymmetric AdaBoost and SVM. The performance of
                  the candidate detector is: sensitivity 95{\%} with
                  5162 FPs/scan. After classification we reduce the
                  number of FPs to 33 for SVM and 41 for asymmetric
                  AdaBoost, while the sensitivity drops only a little
                  to 87.5{\%} and 88{\%} for SVM and asymmetric AdaBoost
  keywords =     {Computer aided diagnosis, nodule detection, matematical
                  morphology, 3D blob detector, SVM, Asymmetric AdaBoost, Lung
  comment =      { Abstrakt ani pocet stran a obrazku jeste neni konecny,
                  aktualizuji je jen co budu mit finalni verzi.},