Evaluation of the models generated from clinical features and deep learning-based segmentations: Can thoracic CT on admission help us to predict hospitalized COVID-19 patients who will require intensive care?

Gülbay, Mutlu; Baştuğ, Aliye; Özkan, Erdem; Öztürk, Büşra Yüce; Mendi, Bökebatur Ahmet Raşit; Bodur, Hürrem

doi:10.1186/s12880-022-00833-2

Table 4 Radiomic parameters, comparison between the groups and results of ROC analysis

From: Evaluation of the models generated from clinical features and deep learning-based segmentations: Can thoracic CT on admission help us to predict hospitalized COVID-19 patients who will require intensive care?

Parameter	Group 1	Group 2	p	Cut-off	Sensitivity %	Specificity %	AUC^c
First order
10th percentile*	− 126.15 (109.66)	− 86.95 (113.60)	0.010^a	> − 87.04	50.0	73.5	0.611
90th percentile*	508.17 ± 147.39	591.50 ± 120.42	< 0.001^b	> 488.26	85.1	45.3	0.672
Energy (× 10¹¹)	5.71 (6.94)	11.4 (9.94)	< 0.001^a	> 6.90	66.2	65.8	0.675
Entropy	5.15 (0.32)	5.29 (0.20)	< 0.001^a	> 5.25	66.2	65.0	0.672
Interquartile Range*	355.01 ± 93.75	398.27 ± 84.74	0.001^a	> 319.42	87.8	40.2	0.644
Kurtosis	2.64 (1.17)	2.26 (0.63)	< 0.001^a	≤ 2.56	74.3	53.0	0.652
Maximum*	998.36 ± 122.10	1009.82 ± 121.38	0.527^b
Mean Absolute Deviation*	199.58 ± 36.68	214.74 ± 32.06	0.003^b	> 201.11	75.7	50.4	0.629
Mean*	162.99 ± 119.13	243.10 ± 122.06	< 0.001^b	> 159.23	77.0	54.7	0.691
Median*	129.73 ± 109.82	231.95 ± 146.89	0.001^b	> 185.67	60.8	71.8	0.702
Minimum*	− 479.49 ± 83.67	− 487.67 ± 89.95	0.506^b
Range*	1477.85 ± 101.11	1497.49 ± 134.92	0.278^b
RMAD *	146.78 ± 35.51	162.46 ± 31.45	0.002^b	> 134.00	86.5	39.3	0.636
Root Mean Squared*	1211.68 ± 121.50	1293.10 ± 121.19	0.001^b	> 1213.80	74.3	57.3	0.693
Skewness	0.578 (0.730)	0.187 (0.648)	< 0.001^a	≤ 0.788	94.6	39.3	0.721
Total Energy (× 10¹¹)	7.09 (6.95)	10.00 (9.94)	< 0.001^a	> 6.9	66.2	65.8	0.675
Uniformity	0.0323 (0.0100)	0.0281 (0.0058)	< 0.001^a	≤ 0.0292	62.2	67,5	0.678
Variance	59,791 ± 17,619	66,375 ± 15,803	0.010^b	> 59,734	73.0	53.0	0.618
Second order—gray level cooccurrence matrix (GLCM)
Autocorrelation	721.06 (420.12)	1009.76 (446.41)	< 0.001^a	> 706.42	87.8	47.9	0.698
Cluster tendency	310.78 (144.06)	354.96 (89.13)	0.012^a	> 311.05	75.7	51.3	0.608
Clustershade	2585.01 (3335.90)	807.02 (4056,87)	< 0.001^a	≤ 1429.10	60.8	70.1	0.676
Maximum probability	0.00464 (0.00345)	0.00353 (0.00332)	0.005^a	≤ 0.00330	47.3	73.5	0.620
Sumentropy	6.0181 (0.3201)	6.1410 (0.1770)	< 0.001^a	> 6.0286	78.4	53.9	0.666
Joint energy (× 10^–3)	1.628 (1.022)	1.320 (0.701)	0.002^a	≤ 1.600	71.6	53.0	0.631
Joint entropy	9.814 (0.514)	10.016 (0.481)	0.005^a	> 9.9209	62.2	62.4	0.620
Correlation	0.6750 ± 0.0810	0.6906 ± 0.0844	0.206^b
Contrast	58.517 (24.270)	61.059 (24.522)	0.203^a
IDN	0.9190 ± 0.0168	0.9169 ± 0.0116	0.238^b
IMC	0.7626 (0.1008)	0.7949 (0.1031)	0.027^a	> 0.8311	35.1	85.5	0.596
Second order—gray level run length matrix (GLRLM)
Gray Level NUN (× 10^–2)	3.202 (0.992)	2.776 (0.577)	< 0.001^a	≤ 2.907	63.5	66.7	0.680
Gray Level Variance	94.812 (39.683)	107.492 (25.276)	0.007^a	> 96.082	73.0	53.0	0.617
High GL Run Emphasis	751.038 (415.427)	1021.780 (440.303)	< 0.001^a	< 724.968	89.2	47.9	0.699
Long-run emphasis	1.192 (0.052)	1.188 (0.058)	0.465^a
Long-run High GLE	882.784 (474.590)	1241.118 (520.097)	< 0.001^a	> 1052.402	68.9	67.5	0.704
Long-run Low GLE(× 10^–3)	2.889 (1.573)	2.103 (1.366)	< 0.001^a	≤ 2.504	67.6	64.1	0.667
Low GL Run Emphasis (× 10^–3)	2.402 (1.146)	1.802 (1.043)	< 0.001^a	≤ 2.107	68.9	63.2	0.673
Run Entropy	5.501 (0.253)	5.618 (0.171)	< 0.001^a	> 5.510	78.4	53.0	0.686
Run Length Nonuniformity	0.892 (0.240)	0.895 (0.269)	0.478^a
Run Percentage	0.942 (0.138)	0.943 (0.154)	0.473^a
Run Variance	0.660 (0.183)	0.646 (0.222)	0.455^a
Short Run Emphasis	0.957 (0.103)	0.958 (0.115)	0.475^a
Short-run High GLE	719.368 (390.961)	982.643 (415.463)	< 0.001^a	> 693.117	89.2	46.2	0.697
Short-run Low GLE (× 10^–3)	2.294 (1.096)	1.730 (0.997)	< 0.001^a	≤ 1.993	68.9	64.1	0.675
Second order—neighborhood gray tone difference matrix (NGTDM)
Busyness	27.061 (27.856)	35.585 (28.361)	0.071^a
Coarseness (× 10^–5)	2.331 (2.957)	1.785 (1.479)	0.007^a	≤ 2.505	75.7	47.9	0.616
Complexity	5736.34 ± 1210.86	6093.04 ± 1887.26	0.150^b
Contrast	0.2082 ± 0.0766	0.2329 ± 0.0694	0.026^b	> 0.2012	66.2	52.1	0.605
Strength (× 10^–2)	3.518 (3.577)	2.365 (1.710)	< 0.001^a	≤ 2.629	63.5	66.7	0.655
Second order—gray level size zone matrix (GLSZM)
GL NUN (× 10^{– 2})	2.810 (0.567)	2.615 (0.299)	< 0.001^a	≤ 2.748	73.0	55.6	0.663
GL Variance	103.41 ± 21.82	110.81 ± 20.47	0.020^a	> 97.76	79.7	43.6	0.610
High GL Zone Emphasis	816.41 (376.85)	1057.04 (340.56)	< 0.001^b	> 873.41	77.0	59.0	0.700
Low GL Zone Emphasis(× 10^–2)	2.429 (1.146)	1.775 (0.880)	< 0.001^a	≤ 2.234	74.3	56.4	0.677
Size Zone NUN	0.4399 (0.0395)	0.4425 (0.0499)	0.884^b
Large Area High GLE	16,758 (42,765)	25,675 (86,414)	0.033^a	> 11,560	85.1	37.6	0.592
Large Area Low GLE	0.0964 (0.3851)	0.0635 (0.2388)	0.229^a
Small Area High GLE	620.68 ± 186.50	745.62 ± 203.19	< 0.001^b	> 616.24	74.3	58.1	0.689
Small Area Low GLE (× 10^–3)	1.642 (0.790)	1.231 (0.627)	< 0.001^a	≤ 1.773	87.8	41.9	0.675
Zone Entropy	7.1386 ± 0.1490	7.2035 ± 0.1828	0.008^b	> 7.2178	59.5	71.8	0.639
Zone Percentage	0.4554 ± 0.0678	0.4623 ± 0.0806	0.527^b
Zone Variance	26.686 (86.459)	29.728 (160.183)	0.981^a
Second order—gray leve dependence matrix (GLDM)
Dependence Entropy	7.410 ± 0.179	7.511 ± 0.194	< 0.001^b	> 7.489	62.2	69.2	0.666
Dependence Nonuniformity	85,241 (96,288)	138,809 (115,232)	0.001^a	> 94,982	74.3	55.6	0.646
Dependence NUN	0.2321 (0.0407)	0.2359 (0.0477)	0.314^b
Dependence variance	2.077 (0.815)	1.983 (0.776)	0.281^a
Dependence GLNU	13,166 (15,270)	17,013 (18,239)	0.031^a	> 15,837	58.1	59.8	0.593
GL Variance	95.750 ± 28.192	106.284 ± 25.285	0.010^b	> 95.667	73.0	53.0	0.618
High GLE	834.296 (424.553)	1047.474 (442.943)	< 0.001^a	> 719.576	89.2	47.9	0.698
Low GLE (× 10^–3)	2.408 (1.157)	1.801 (1.059)	< 0.001^a	≤ 2.126	68.9	63.2	0.672
Large Dependence Emphasis	8.245 (2.248)	8.690 (2.938)	0.419^a
Small Dependence Emphasis	0.397 ± 0.055	0.403 ± 0.067	0.534^b
Large DHGLE	5541.21 (2636.73)	7369.43 (3673.84)	< 0.001^a	> 6562.75	70.3	58.4	0.708
Large DLGLE (× 10^{– 2})	1.979 (2.010)	1.388 (1.199)	0.001^a	≤ 1.455	59.5	68.4	0.646
Small DHGLE	354.751 ± 134.628	435.646 ± 147.741	< 0.001^b	> 331.553	78.4	53.0	0.670
Small DLGLE (× 10^{– 4})	8.898 (4.828)	6.843 (3.277)	< 0.001^a	≤ 8.198	68.9	62.4	0.678

Group results were given as Median (IQR) or Mean ± SD according to the parameter’s Nonnormal or Normal distribution, respectively
Group1: Patients discharged from inpatient floor; Group 2: Patients transferred to ICU
IDN inverse difference normalized, IMC informational measure of correlation, RMAD robust mean absolute deviation, GL gray level, GLE gray level emphasis, NUN nonuniformity normalized, DHGLE dependence high gray level emphasis, DLGLE dependence low gray level emphasis
^*Standardized results
^aMann–Whitney test result
^bT-test result
^cROC analysis result

Back to article page

ISSN: 1471-2342

Contact us

General enquiries: journalsubmissions@springernature.com

BMC Medical Imaging

Contact us