Table 5.

Characterization using DL

Anatomic site	Object or task	Network input	Network architecture	Dataset (train/test)
Breast	Cancer risk assessment192	Mammograms	Pretrained Alexnet followed by SVM	456 patients LOO CV
	Cancer risk assessment193	Mammograms	Modified AlexNet	14,000/1850 images randomly selected 20 times
	Cancer risk assessment194	Mammograms	Custom DCNN	478/183 mammograms
	Cancer risk assessment195	Mammograms	Fine‐tuned a pretrained VGG16Net	513/91 women
	Diagnosis196	Mammograms	Pretrained AlexNet followed by SVM	607 cases fivefold CV
	Diagnosis197	Mammograms, MRI, US	Pretrained VGG19Net followed by SVM	690 MRI, 245 FFDM 1125 US, LOO CV
	Diagnosis198	Breast tomosynthesis	Pretrained Alexnet followed by evolutionary pruning	2682/89 masses
	Diagnosis199	Mammograms	Pretrained AlexNet	1545/909 masses
	Diagnosis200	MRI MIP	Pretrained VGG19Net followed by SVM	690 cases with fivefold CV
	Diagnosis201	DCE‐MRI	LSTM	562/141 cases
	Solitary cyst diagnosis202	Mammograms	Modified VGG Net	1600 lesions eightfold CV
	Prognosis203	Mammograms	VGG16Net followed by logistic regression classifier	79/20 cases randomly selected 100 times
Chest — lung	Pulmonary nodule classification204	CT patches	ResNet	665/166 nodules
	Tissue classification205	CT patches	Restricted Boltzmann machines	Training 50/100/150/200; testing 20,000/1000/20,000/20,000 image patches
	Interstitial disease206	CT patches	Modified AlexNet	100/20 patients
	Interstitial disease207	CT patches	Modified VGG	Public: 71/23 scans Local: 20/6 scans
	Interstitial disease208	CT patches	Custom	480/(120 and 240)
	Interstitial disease209	CT patches	Custom	36,106/1050 patches
	Pulmonary nodule staging210	CT	DFCNet	11/7 patients
	Prognosis211	CT	Custom	7983/(1000 and 2164) subjects
Chest — cardiac	Calcium scoring212	CT	Custom	1181/506 scans
	Ventricle quantification213	MR	Custom (CNN + RNN + Bayesian multitask)	145 cases, fivefold CV
Abdomen	Tissue classification214	Ultrasound	CaffeNet and VGGNet	136/49 Studies
	Liver tumor classification215	Portal Phase 2D CT	GAN	182 cases, threefold CV
	Liver Fibrosis216	DCE‐CT	Custom CNN	460/100 scans
	Fatty liver disease217	US	Invariant scattering convolution network	650 patients, five‐ and tenfold CV
Brain	Survival218	Multiparametric MR	Transfer learning as feature extractor, CNN‐S	75/37 patients
Skeletal	Maturity219	Hand radiographs	Deep residual network	14,036/(200 and 913) examinations

FCN, fully convolutional network; LOO, leave‐one‐out; CV, cross‐validation.