Table 1.
Introduction of datasets.
Dataset | Number of CXRs | Description | Application |
---|---|---|---|
VGHTPE training | 559 | 659 CXRs with biopsy confirmed lung nodules obtained from VGHTPE and separated into 559 CXRs for training dataset and 100 CXRs isolated for testing dataset. | Training |
VGHTPE testing | 100 | Testing | |
Shenzhen [13] | 662 | 662 CXRs with segmented lung masks manually delineated by Stirenko and colleagues. | Segmentation |
Montgomery [13] | 138 | 80 normal CXRs and 58 CXRs with tuberculosis manifestations. All with lung masks. | Segmentation |
NIH [14] | 880 | 880 CXRs with annotations in 112,120 CXRs from NIH open dataset | Pretrain |
JSRT [15] | 154 | 154 CXRs with lung nodules in 247 CXRs collected by JSRT. | Testing |
JSRT—the Japanese Society of Radiological Technology; NIH—National Institutes of Health; VGHTPE—Taipei Veterans General Hospital.