A Pneumonia Diagnosis Scheme Based on Hybrid Features Extracted from Chest Radiographs Using an Ensemble Learning Algorithm

Table 5

Performance comparison with other similar methods.

Reference	Identified classes	Number of samples	Classification model	Accuracy (%)	F-score (%)	Specificity (%)
[1]	N–P	5856	Multichannel CNN	97.92	97.97	—
[8]	N–P	5232	Inception V3	92.80	—	90.10
[8]	B–V	3883	Inception V3	90.70	—	90.90
[10]	N–P	5232	Ensemble of multiple CNNs	96.39	96.35	—
[11]	N–P	5857	Ensemble of convolutions with capsules	95.90	—	—
[12]	N–P	5856	PneumoniaNet	94.80	95.90	96.70
[13]	N–P	112, 120	MD-Conv	93.40	—	86.80
[13]	B–V	—	MD-Conv	91.00	—	92.60
[14]	N–P	5856	Custom CNN	93.73	—	—
[15]	H–B–V	6161	CovXNet	91.70	92.60	93.60
[15]	H–B–V	6161	ResidualNet	86.30	87.40	93.50
[15]	H–B–V	6161	InceptionNet	81.10	78.90	86.20
[15]	H–B–V	6161	VGG-19	79.80	77.90	83.40
[16]	H–B–V	5840	Model 1	85.26	89.00	—
[16]	H–B–V	5840	Model 2	92.31	94.00	—
[16]	H–B–V	5840	ResNet50	77.56	—	—
[16]	H–B–V	5840	Inception-v3	70.99	—	—
Proposed	H–B–V	4672	CNN + RandomForest	86.30	86.03	95.60

“--” denotes that the information is not mentioned in the associated study.