Читать книгу Computational Intelligence and Healthcare Informatics - Группа авторов - Страница 46

In this section, existing models are compared on many different parameters as listed in Table 2.2.

1 Model parameters: type of model used, input image size, number of layers epoch, loss function used, and accuracy.

2 Accuracy achieved for all 14 different pathologies and dataset used for experimentation.

3 Other metrics: model used, specificity, sensitivity, F1-score, precision, and type of pathology detected.

4 On the basis of hardware and software used and input image size.

Table 2.2 shows datasets and pathologies detected by various DL models. It is observed that ChestX-ray14 is the most preferred dataset which can be used for training and testing of newly implemented models in the future. Cardiomegaly and Edema are easy to detect as compared to other pathologies because of their spatially spread out symptom. In addition, Table 2.3 shows comparison among different models on the basis of AUC score for 14 different chest pathologies. It is observed that cardiomegaly is having the highest value of AUC score obtained by most ensemble models. Edema, Emphysema, and Hernia are the pathologies following cardiomegaly which are having better AUC. CheXNext [59] is able to detect Mass, Pneumothorax, and Edema more accurately which other models cannot detect.

Table 2.4 highlights the comparison between various models on the basis of other performance measure. Most of the models have used accuracy, F1-score, specificity, sensitivity, and PPV and NPV parameters for comparison with other models. Pre-trained networks like AlexNet, VGG16, VGG19, DenseNet, and ResNet121 trained from scratch achieved better accuracy than those whose parameters are initialized from ImageNet because ImageNet has altogether different features than CXR images.

Table 2.5 shows hardware used by different models along with size of input image in terms of pixels and datasets used. Due to computationally intensive task, high definition hardware of NVIDIA card with larger size RAM has been used.

Table 2.2 Comparison of different deep learning models.

Ref.	Model used	Dataset	No. layers	Epoch	Activation function	Iterations	Pathology detected
[23]	DenseNet-121	ChestX-ray14	121	-	Softmax	50,000	14 chest pathologies
[67]	Pretrained CNNs:	ChestX-ray14	-	50	-	-	14 chest pathologies
[7]	VDSNet	ChestX-ray8	-	-	ReLU	-	Pulmonary diseases
[10]	DualCheXNet	ChestX-ray14	169	-	ReLU	-	14 chest pathologies
[2]	CNN	ChestX-ray8	5	1,000–4,000	ReLU	40,000	12 chest pathologies out of 14
BPNN	3		-	5,000
CpNN	2		-	1,000
[8]	Customized U-Net	ChestX-ray8	35	100–200	ReLU	20	Cardiomegaly
[47]	Ensemble of DesnSeNet-121, DenseNet-169, DenseNet-201, Inception-ResNet-v2 Xception, NASNetLarge	CheXpert	5		Sigmoid	50,000	Only 5 pathologies: Atelactasis, Cardiomegaly, Pleural, Effusion, and Edema
[51]	STN based CNN	Lung ultrasonography videos	-	-	ReLU	-	COVID-19 Pneumonia
[46]	Ensemble with AlexNet	NIH Tuberculosis Chest X-ray dataset [10.31] and Belarus Tuberculosis Portal dataset [10.32]	-	-	ReLU	-	Tuberculosis
[26]	FHRNet	ChestX-ray14 dataset	-	50	Sigmoid	-	14 chest pathologies
[20]	AG-CNN	ChestX-ray14		50	Sigmoid	-	14 chest pathologies
[53]	preact-ResNet [39_chp7_14]	ChestX-ray14	-	-	Sigmoid	-	8 chest pathologies out of 14
[70]	Unified DCNN	ChestX-ray8	-	-	Sigmoid	-	8 chest pathologies out of 14
[28]	Ensemble of AlexNet, VGG-16, VGG-19, ResNet-50, ResNet-101 and ResNet-152	Indiana dataset	-	-	-	-	Cardiomegaly, Edema, and Tuberculosis
[5]	ChestNet CNN	ChestX-ray14	14	30			Consolidation
[49]	CheXNeXt CNN	ChestX-ray14					14 chest pathologies
[46]	Customized CNN	NIH Tuberculosis Chest X-ray, Belarus Tuberculosis	23		ReLU		Tuberculosis
[61]	Ensemble of RetinaNet and Mask R-CNN	Kaggle dataset RSNA	-	-	ReLU	-	Pneumonia
[45]	DarkCovid-Net	COVID Chest x-ray Kaggle	17		ReLU	-	COVID, Normal and Pneumonia
[65]	GoogLeNet	St. Michael’s Hospital chest x-ray	-	60	ReLU	-	5 pathologies: Cardiomegaly, Edema, Pleural effusion, pneumothorax, and consolidation
[35]	Ensemble of AlexNet and GoogleNet	NIH Tuberculosis Chest X-ray, Belarus Tuberculosis	-	-	-	-	Tuberculosis
[31]	ResNet101	Cohen and Kaggle [3]	-	-	-	-	COVID-19–induced pneumonia

Table 2.3 Comparison of models on the basis of AUC score for 14 chest pathologies.

Ref.	Atel	Card	Effu	Infi	Mass	Nodu	Pne1	Pnet	Cons	Edem	Emph	Fibr	PT	Hern
[10]	0.784	0.888	0.831	0.705	0.838	0.796	0.727	0.876	0.746	0.852	0.942	0.837	0.796	0.912
[23]	0.795	0.887	0.875	0.703	0.835	0.716	0.742	0.863	0.786	0.892	0.875	0.756	0.774	0.836
[70]	0.716	0.807	0.784	0.609	0.706	0.671	0.633	0.806	0.708	0.835	0.815	0.769	0.708	0.767
[69]	0.743	0.875	0.811	0.677	0.783	0.698	0.696	0.81	0.723	0.833	0.822	0.804	0.751	0.899
[8]	-	0.940	-	-	-	-	-	-	-	-	-	-	-	-
[47]	0.909	0.910	0.964	-	-	-	-	-	0.957	0.958	-	-	-	-
[26]	0.794	0902	0.839	0.714	0.827	0.727	0.703	0.848	0.773	0.834	0.911	0.824	0.752	0.916
[20]	0.853	0.939	0.903	0.754	0.902	0.828	0.774	0.921	0.842	0.924	0.932	0.964	0.837	0.921
[74] AlexNet	0.645	0.692	0.664	0.604	0.564	0.648	0.549	0.742	-	-	-	-	-	-
GoogleNet	0.631	0.706	0.688	0.609	0.536	0.558	0.599	0.782	-	-	-	-	-	-
VGGNet-16	0.628	0.708	0.650	0.589	0.510	0.656	0.510	0.751	-	-	-	-	-	-
ResNet50	0.707	0.814	0.736	0.613	0.561	0.716	0.63	0.789	-	-	-	-	-	-
[41]	0.762	0.883	0.816	0.679	0.801	0.729	0.709	0.838	0.744	0.841	0.884	0.800	0.754	0.876
[49]	0.862	0.831	0.901	0.721	0.909	0.894	0.851	0.944	0.893	0.924	0.704	0.806	0.798	0.851
[13]	-	0.875	0.962	-	-	-	-	0.861	0.850	0.868	-	-	-	-

Table 2.4 Comparison of DL models on the basis of different performance metrics.

Ref.	Model	Accuracy	F1-Score	Specificity	Sensitivity	PPV	NPV	AUC	Recall	Precision
[23]	DenseNet121	0.572–0.842	0.574 –0.942	-	-	-	-	-	-	-
[67]	AlexNet (S)	0.9684	0.9023	87.99	92.65	87.94	90.68	-	-	-
VGG16 (S)	0.9742	0.9228	91.46	93.42	91.18	93.63	-	-	-
VGG19 (S)	0.9757	0.9161	88.86	94.49	88.90	94.46	-	-	-
DenseNet121(S)	0.9801	0.9248	90.01	95.10	90.00	95.11	-	-	-
ResNet18 (S)	0.9766	0.9099	85.09	96.63	85.97	96.52	-	-	-
Inceptionv3(S)	0.9796	0.9225	89.58	95.08	89.58	95.08	-	-	-
ResNet50 (S)	0.9775	0.9233	90.59	94.32	90.43	94.42	-	-	-
[7]	VDSNet	0.73	0.68	-	-	-	-	0.74	0.63	0.69
[10]	DualCheXNet	-	-	-	-	-	-	0.823	-	-
[2]	CNN	0.9240	-	-	-	-	-	-	-	-
BPNN	0.8004	-	-	-	-	-	-	-	-
CpNN	0.8957	-	-	-	-	-	-	-	-
[8]	U-Net	0.94	-	-	-	-	-	-	-	-
[47]	Ensemble CNN							0.940
[51]	STN based CNN	0.96	0.651	-	-	-	-	-	0.60	0.70
[46]	Ensemble with AlexNet	0.862						0.925
[26]	FHRNet	-	-	-	-	-	-	0.812	-	-
[20]	AG-CNN	-	-	-	-	-	-	0.871	-	-
[28]	Ensemble DCNN (for Cardiomegaly)	0.93	-	92.00	94.00	-	-	0.97	-	-
Ensemble DCNN (for Tuberculosis)	0.90	-	92.00	88.00	-	-	0.94	-	-
[41]	MA-DCNN	-	-	-	-	-	-	0.794	-	-
[5]	ChestNet	0.932	-	-	97.13	85.85	-	0.984	-	-
[46] CNN	Maryland (MC) dataset	0.79	-	-	-	-	-	0.811	-	-
Shenzhen (SZ)	0.844	-	-	-	-	-	0.900	-	-
Combined (CB)	0.862	-	-	-	-	-	0.925	-	-
[61]	Kaggle PSNA	-	0.755	-	-	-	-	-	0.793	0.758
[45]	DarkCovidNet	0.981	94.17	99.61	90.65	-	-	-	-	0.979
[13]	GoogleNet (for normal class)	-	-	91.00	91.00	-	-	0.964	-	-
[4]	DecafCNN [78]	-	-	78.00	84.00	-	-	87.00	-	-
[35]	Ensemble DCNN	-	-	-	-	-	-	0.99	-	-
[31]	ResNet101	0.989	98.15	98.66	98.93	-	-	-	-	0.964

Table 2.5 Models with hardware used and time required for training.

Ref.	Dataset	Hardware and software platform used	Input image size	Time required for training
[47]	CheXpert	NVIDIA Geforce RTX 2080 Ti with 11GB memory. Python with Keras and TensorFlow	224 × 224 pixels	-
[51]	Lung ultrasonography videos from Italy	RTX-2080 NVIDIA GPU	1,005 frames	11 hours
[46]	NIH Tuberculosis Chest X-ray dataset [18] and Belarus Tuberculosis Portal dataset [21]	Nvidia GeForce GTX 1050 Ti	512 × 512	5–6 ms
[26]	ChestX-ray14 dataset	8-core CPU and four TITAN V GPUs Pytorch 1.0 framework in Python 3.6 on an Ubuntu 16.04 server	224 × 224	-
[20]	ChestX-ray14 dataset	NVIDIA TITAN Xp GPUs Pytorch	224 × 224	6 hours
[70]	ChestX-ray14 dataset	Dev-Box linux server with 4 Titan X GPUs	224 × 224	-
[5]	ChestX-ray14 dataset	Intel Core(TM) i7-6850k CPU 3.60GHz processor, 4TB of hard disk space, 7889 MB of RAM, and a CUDA-enabled NVidia Titan 11 GB graphics processing unit with python and Keras library on TensorFlow	224 × 224	-
[49]	ChestX-ray8	NVIDIA GeForce GTX TITAN and PyTorch	512 × 512	20 hours
[46]	NIH Tuberculosis Chest X-ray [18], Belarus Tuberculosis [A6]	Nvidia GeForce GTX 1050 Ti	512 × 512	1 hour
[61]	Kaggle PSNA	Nvidia Tesla V100 and Nvidia K80 and Keras library of Python	512 × 512	7 hours
[13]	St. Michael’s Hospital chest x-ray	3 NVIDIA Titan X 12GB GPUs	256 × 256	1 hour
[35]	NIH Tuberculosis Chest X-ray [18], Belarus Tuberculosis [A6]	Intel i5 processor with 32 GB of RAM, and a CUDA-enabled Nvidia Titan 312 GB GPU	256 × 256	-