Skip to main content

Table 1 The summary of the feature set we designed for predicting NRM

From: Prenatal prediction of neonatal respiratory morbidity: a radiomics method based on imbalanced few-shot fetal lung ultrasound images

Feature type

Feature name

Feature number

Clinical information

(1) GA, (2) GDM

2

Greyscale histogram features

(3) Energy, (4) Entropy, (5) Kurtosis, (6) Mean, (7) Median absolute deviation, (8) Median, (9) Range, (10) Uniformity, (11) Variance, (12) Root mean square, (13) Skewness, (14) Deviation, (15) Histogram kurtosis, (16) Histogram mean, (17) Histogram variance, (18) Histogram skewness

16

ROI textural features

(19) Mean of contrast, (20) SD of contrast, (21) Mean of covariance, (22) SD of covariance, (23) Mean of non-similarity, (24) SD of non-similarity

6

GLCM textural features

(25) Energy, (26) Entropy, (27) Dissimilarity, (28) Contrast, (29) Inversed difference, (30) Correlation 1, (31) Correlation 2, (32) Homogeneity, (33) Autocorrelation, (34) Cluster shade, (35) Cluster prominence, (36) Maximum probability, (37) Sum of squares, (38) Sum average, (39) Sum variance, (40) Sum entropy, (41) Difference variance, (42) Difference entropy, (43) Information measures of correlation 1, (44) Information measures of correlation 2, (45) Maximal correlation coefficient, (46) Inverse difference normalized, (47) Inverse difference moment normalized

23

GLRLM textural features

(48) Short-run emphasis, (49) Long-run emphasis, (50) Grey-level non-uniformity, (51) Run length non-uniformity, (52) Run percentage, (53) Low grey-level run emphasis, (54) High grey-level run emphasis, (55) Short-run low grey-level emphasis, (56) Short-run high grey-level emphasis, (57) Long-run low grey-level emphasis, (58) Long-run high grey-level emphasis, (59) Grey-level variance, (60) Run-length variance

13

GLSZM textural features

(61) Small zone emphasis, (62) Large zone emphasis, (63) Grey-level non-uniformity, (64) Zone size non-uniformity, (65) Zone percentage, (66) Low grey-level zone emphasis, (67) High grey-level zone emphasis, (68) Small zone low grey-level emphasis, (69) Small zone high grey-level emphasis, (70) Large zone low grey-level emphasis, (71) Large zone high grey-level emphasis, (72) Grey-level variance, (73) Zone-size variance

13

NGTDM textural features

(74) Coarseness, (75) Contrast, (76) Busyness, (77) Complexity, (78) Strength

5

Wavelet features

(79–154) Approximation, (155–230) Horizontal, (231–306) Vertical, (307–382) Diagonal

304

Total feature number

 

382

  1. (1) Clinical information: GA and GDM are strongly correlated with NRM [7, 8]. GA was determined by the last menstrual period and verified by first-trimester dating ultrasound (crown-rump length). According to the presence of GDM during pregnancy, these pregnant women were divided into Yes and No groups
  2. (2) Greyscale histogram features: Describe the greyscale and histogram distribution of the ROI in fetal lung ultrasound images [13]
  3. (3) Textural features: Describe detailed, invisible greyscale changes and associations in fetal lung ultrasound images
  4. (a) ROI textural features: Describe the distribution of greyscale inside the ROI [14]
  5. (b) Grey-level co-occurrence matrix (GLCM) textural features: Describe the specified spatial linear relationship between the frequencies of two greyscale intensities inside the ROI [15]
  6. (c) Grey-level run-length matrix (GLRLM) textural features: Describe the roughness of the texture by calculating the run-length of the collinear image pixels of the same grey-level in a given direction inside the ROI [16, 17]
  7. (d) Grey-level size zone matrix (GLSZM) textural features: Describe the uniformity of the small pixel population of the ROI [15, 18]
  8. (e) Neighbourhood grey-tone difference matrix (NGTDM) textural features: Describe the difference between the greyscale of each image pixel and the greyscale of its neighbours inside the ROI [19]
  9. (4) Wavelet features: Describe information that is not directly reflected by the greyscale and textural features of the original image. Every fetal lung ultrasound image was decomposed into four components: approximate, horizontal, vertical, and diagonal by wavelet transform (first-level decomposition). Then, the 76 features mentioned above were extracted separately on each component. Finally, a total of 304 wavelet features were extracted
  10. Approximate, horizontal, vertical, and diagonal were decomposed from the image by wavelet transform (first-level decomposition)
  11. GA: gestational age, GDM: gestational diabetes mellitus, ROI: region of interest (fetal lung region), SD: standard deviation, GLCM: grey-level co-occurrence matrix, GLRLM: grey-level run-length matrix, GLSZM: grey-level size zone matrix, NGTDM: neighbourhood grey-tone difference matrix
Back to article page

BMC Medical Imaging

ISSN: 1471-2342

Contact us