Skip to main content

Table 2 Predictive performance of significant clinical variables, the deep learning signature, and clinical and clinical-deep learning models in classification of WDLPS and lipoma on patients in the training and validation cohorts

From: Novel computer aided diagnostic models on multimodality medical images to differentiate well differentiated liposarcomas from lipomas approached by deep learning methods

Parameters

Training cohort

Validation cohort

AUC

Accuracy

Sensitivity

Specificity

PPV

NPV

AUC

Accuracy

Sensitivity

Specificity

PPV

NPV

Variables

            

Age

0.625 (0.505–0.746)

66.29 (59/89)

36.84 (22.29–54.00)

88.24 (75.44–95.13)

70.00 (45.67–87.16)

65.22 (52.71–76.02)

0.514 (0.328–0.700)

50.00 (19/38)

25.00 (9.59–49.41)

77.78 (51.92–92.63)

55.56 (22.65–84.66)

48.28 (29.89–67.10)

LDH

0.572 (0.450–0.695)

62.92 (56/89)

18.42 (8.32–34.89)

44.09 (85.41–99.32)

77.78 (40.19–96.05)

61.25 (49.67–71.74)

0.436 (0.251–0.622)

42.11 (16/38)

15.00 (3.96–38.86)

72.22 (46.41–89.29)

37.50 (10.24–74.11)

43.33 (25.98–62.34)

DL signaturea

0.995 (0.987–1.000)

95.51 (85/89)

92.11 (77.52–97.94)

98.04 (88.21–99.90)

97.22 (83.80–99.85)

94.34 (83.37–98.53)

0.950 (0.886–1.000)

92.11 (35/38)

95.00 (73.06–99.74)

88.89 (63.93–98.05)

90.48 (68.17–98.33)

94.12 (69.24–99.69)

Models

            

Clinical modelb

0.652 (0.534–0.770)

65.17 (58/89)

39.47 (24.49–56.55)

84.31 (70.86–92.52)

65.22 (42.82–82.81)

65.15 (52.34–76.19)

0.504 (0.318–0.690)

50.00 (19/38)

40.00 (19.98–63.59)

61.11 (36.14–81.74)

53.33 (27.42–77.72)

47.83 (27.42–68.92)

Clinical-DL modelc

0.996 (0.989–1.000)

95.51 (85/89)

97.37 (84.57–99.86)

94.12 (82.77–98.47)

92.50 (78.52–98.04)

97.96 (87.76–99.89)

0.942 (0.867–1.000)

86.84 (33/38)

95.00 (73.06–99.74)

77.78 (51.92–92.63)

82.61 (60.45–94.28)

93.33 (66.03–99.65)

  1. WDLPS, Well-differentiated liposarcoma; LDH, Lactate dehydrogenase; DL, Deep learning; AUC, Area under the receiver operating characteristic curve; PPV, Positive predictive value; NPV, Negative predictive value
  2. aThe deep learning signature was generated from the best deep learning model considering AUC of the validation cohort on different imaging examinations
  3. bThe clinical model was constructed by significant clinical variable, age at diagnosis and LDH selected by multivariate analysis with p value less than 0.05 in the training cohort
  4. cThe clinical-deep learning model was constructed by significant clinical variables, age at diagnosis and LDH, and the deep learning signature
Back to article page

Orphanet Journal of Rare Diseases

ISSN: 1750-1172