Table 3 Comparison of classification accuracy (%) across different models and training methods (Single input vs multi input) for retinal disease image classification.

From: A deep learning model for diagnosis of inherited retinal diseases

Algorithm		Input	Accuracy	Balanced accuracy	Precision	F1	Specificity	NPV	AUC
ML algorithms	XGBoost ²⁸	CFP	73.46	63.86	59.9	64.83	81.83	92.42	85.13
		IR	73.46	63.86	59.9	64.83	81.83	92.42	85.13
		Multi-Input	75.93 (74.69–76.55)	66.16	59.97	66.51	83.77	93.84	87.15
	LightGBM ³⁰	CFP	73.46	63.86	58.8	64.49	82.07	92.42	87.69
		IR	75.31	71.26	74.69	74.01	85.63	88.42	91.94
		Multi-Input	77.16 (75.3-78.14)	67.92	77.66	69.81	84.53	94.02	91.17
DL networks	AlexNet ³²	CFP	87.04	83.37	87.13	86.18	92.35	95.06	94.06
		IR	87.04	83.37	87.13	87.04	86.18	92.35	95.06
		Multi-Input	87.65 (87.65–90.12)	86.55	87.83	87.52	94.19	93.95	97.58
	ShuffleNetV2 1 × ³⁴	CFP	90.12	88.97	90.12	90.12	95.26	95.26	98.22
		IR	90.74	88.63	90.78	90.5	94.58	96.07	95.41
		Multi-Input	93.83 (91.36–94.44)	93.09	93.76	93.78	96.86	96.96	98.39
	Inception V3 ³⁵	CFP	92.59	90.22	92.72	92.3	95.54	97.32	98.39
		IR	93.21	92.51	93.27	93.23	96.51	96.57	97.59
		Multi-Input	95.68 (95.06–96.3)	94.6	95.71	95.64	97.5	98.2	99.37
	ResNet50 ³⁶	CFP	93.21	93.33	93.75	93.27	97.37	96.24	98.39
		IR	93.83	91.97	93.91	93.67	96.31	97.66	97.63
		Multi-Input	95.06 (94.44–95.06)	93.73	95.13	95	97.08	98.01	98.25
	VGG16 ³³	CFP	94.44	92.85	94.56	94.35	96.65	97.83	98.83
		IR	92.59	90.59	92.5	92.35	95.93	97.08	97.8
		Multi-Input	95.68 (95.06–96.3)	94.98	95.64	95.65	97.82	97.98	98.52
	DenseNet121 ³⁷	CFP	93.83	92.72	93.76	93.74	96.7	97.19	97.19
		IR	94.44	93.15	94.38	94.37	97.05	97.68	98.83
		Multi-Input	95.06 (94.44–95.68)	94.47	95.04	95.05	97.63	97.56	98.83
	MobileNetV2 ³⁸	CFP	94.44	93.15	94.38	94.37	97.05	97.68	98.83
		IR	94.44	93.59	94.43	94.4	97.36	97.4	99.15
		Multi-Input	96.3 (95.06–96.3)	95.48	96.3	96.27	97.92	98.39	99.31

Signifiacnce value bold.

Back to article page

Table 3 Comparison of classification accuracy (%) across different models and training methods (Single input vs multi input) for retinal disease image classification.

Search

Quick links