Differential diagnosis of Crohn’s disease and intestinal tuberculosis based on ATR-FTIR spectroscopy combined with machine learning

Table 1 Signature fingerprint of infrared spectra

Bands position (cm-1)	Assignments of group	Assignments of substance
2925	υas, CH3	Lipid related
2855	υas, CH2	Lipid related
1740	υC=O	Lipid
1640	Amide I	Protein
1550	Amide II	Protein
1460	δC-H	Lipid related
1400	δC-H, δC-O-H	Lipid related
1305	δC-H, δC-O-H	Undetermined
1240	υas, PO2-	Nucleic acid related
1160	υC-O, δC-O-H, δC-O-C	Carbohydrate related
1120	υC-O, δC-O-H, δC-O-C	Carbohydrate related
1080	υas, PO2-	Nucleic acid related

Table 2 Sample set divisions of Crohn’s disease and intestinal tuberculosis

Sample set	CD	ITB	Total
Training set	79	66	145
Prediction set	22	27	49

CD: Crohn’s disease; ITB: Intestinal tuberculosis.

Table 3 Results of the XGBoost model

Group	True value	Predicted value		Accuracy (%)
		ITB	CD
Original spectral data	ITB	17.0000	5.0000
	CD	4.0000	23.0000
	Specificity (%)	0.8519
	Sensitivity (%)		0.7727
	Accuracy (%)			0.8163
First derivative spectral data	ITB	20.0000	2.0000
	CD	2.0000	25.0000
	Specificity (%)	0.9259
	Sensitivity (%)		0.9090
	Accuracy (%)			0.9184
Second derivative spectral data	ITB	16.0000	6.0000
	CD	1.0000	26.0000
	Specificity (%)	0.9630
	Sensitivity (%)		0.7270
	Accuracy (%)			0.8571

CD: Crohn’s disease; ITB: Intestinal tuberculosis.

Table 4 Optimal parameters of the XGBoost model based on first-derivative spectral data

Parameters	Step length	Optimal range	AUC (%)		Accuracy (%)
			Optimal value	Test	Optimal value	Test
Max_depth	1.00	(1, 50.0)	3.0	79.9	4.0.	74.5
N_estimators	10.00	-1500	71.0	80.3	81.0	74.4
Min_child_weight	1.00	(1, 30.0)	4.0	82.1	4.0	76.0
Gamma	1.00	(0, 15.0)	0	82.1	0	76.0
Subsample	0.10	(0, 1.1)	1.0.	82.1	1.0.	76.0
Alpha	0.10	(0, 10.0)	0.3	82.0	2.8	75.9
Learning_rate	0.01	(0, 0.2)	0.1	82.0	0.1	75.2

AUC: Area under the curve.