LViT -

• 데이터에 MonuSeg 쓰긴 했는데
• We extend text annotations on the QaTaCOV19 dataset for the first time with the help of professionals.
• The text annotations on MosMedData+
  dataset mainly contain the same information as QaTa-COV19
  dataset, and the text structure is similar
• Those text annotations were provided and verified by two professionals from the Department of Radiation Oncology, UT Southwestern Medical Center. The radiologists independently annotated the same image, and then we compared their annotations to ensure.

tq!! 전문가의 주석이 들어갔음 ㅇ<-<

Three datasets are used in the experiments to evaluate the
performance of our method. The first one is the MosMedData+1
dataset [11], [12], which contains 2729 CT scan slices of lung
infections. The second one is the QaTa-COV19 dataset [13],
which is compiled by researchers from Qatar University and
Tampere University. This dataset consists of 9258 COVID-19
chest X-ray radiographs with manual annotations of COVID19 lesions for the first time. In addition, text annotations for
the datasets are extended by us to be used for training the
vision-language model. We extend text annotations on the QaTaCOV19 dataset for the first time with the help of professionals.
The text annotations focuse on whether both lungs are infected,
the number of lesion regions, and the approximate location
of the infected areas. For example, "Bilateral pulmonary
infection, two infected areas, upper left lung and upper
right lung." refers to bilateral lung infection, and there are
two infection areas located in the upper left lung and the upper
right lung respectively. The text annotations on MosMedData+
dataset mainly contain the same information as QaTa-COV19
dataset, and the text structure is similar, e.g., "Unilateral
pulmonary infection, two infected areas, middle lower
left lung.". The third dataset is the ESO-CT dataset, which
consists of 286 cases, and the detail will be presented in
the section of generalization study. Those text annotations
were provided and verified by two professionals from the
Department of Radiation Oncology, UT Southwestern Medical
Center. The radiologists independently annotated the same
image, and then we compared their annotations to ensur