【醫學影像系列三】青光眼診斷眼底圖像資料集｜代碼｜論文總結｜結果彙總｜名詞解析｜評價名額

導語：這是之前做青光眼診斷研究時候整理的資料，所有的文章數字編号都對應我之前發表的論文閱讀部落格，比如方法彙總 No.15 對應【醫學+深度論文：F15】這篇文章。有一些對大家有用但是我沒有用到的學習資料也放在文中。彙總整理難免有遺漏，希望大家能夠提出和補充，謝謝～

Dataset

資料集彙總

Dataset	Name & Paper	Country	Photo	Feature	Equipment	disease
Drishti-GS	Retinal image dataset for optic nerve head(ONH) segmentation Paper	India	101	OD/OC	FOV 30° 2896×1944 png	Glaucoma
DRIONS-DB	Identification of the optic nerve head with genetic algorithms Paper	110	OD	600×400	Glaucoma
HRF	High-Resolution Fundus (HRF) Image Database	15N，G15，DR15	vessel segment	Canon CR-1 fundus camera FOV 45°	Glaucoma DR
ORIGA	Digital Retinal Images for Vessel Extraction ORIGA-light Paper	650（168G、482N）	OD/OC	3072×2048
RIM-ONE	An open retinal image database for optic nerve evaluation Paper	455 (255 N,200G) 159 (74G,85N)	OD/OC	Nidek AFC210 with a body of a Canon EOS 5D Mark II of 21.1 megapixels 2144×1424
RIM-ONE r1	158( 40 G ,118 N)	OD/OC
RIM-ONE r2	425 (200 G , 225 N)	OD/OC
RIM-ONE r3	159( 74 G, 85 N)	stereo fundus
ONHSD	Optic nerve head segmentation Paper	99 (50 G,49 N)	OD	Canon CR6 45MNf fundus camera FOV 45° 640×480	Glaucoma
SEED-DB
MESSIDOR	Methods to evaluate segmentation and indexing techniques in the field of retinal ophthalmology	1200	DR	DR
SCES	Singapore Chinese Eye Study	1676 (46G,1630N)
SINDI	Singapore Indian Eye Study	5783（5670N，113G）
REFUGE	400 (360N ， 40 G)	macula/OD	Zeiss Visucam 500 2124x2056
GRI	GRI’s Sustainability Disclosure Database
DIARETDB1	DIARETDB1-Standard Diabetic Retinopathy Database Calibration level 1
STARE	Structured Analysis of the Retina
sjchoi86-HRF
ACRIMA
Tajimi	The Tajimi study report 2:prevalence of primary angle closure and secondary glaucoma in a Japanese population	Japan	162 (81 NFLDs , 81 N) 261 (130 NFLDs ,131 N)	NFLD	IMAGEnet digital fundus camera system (TRC-NW6S, Topcon, Tokyo, Japan) 768x576，JPEG
AREDS	Age-Related Eye Disease Study–Results	AMD /cataract

其他

下面是待整理資料集，大家有需要可以自己查，我當時沒有用到這些。

Website. http://www.ukbio- bank.ac.uk/about-biobank-uk. Accessed December 5, 2018

OD

SiMES

INSPIRE-AVR

Shifa

3CHASE-DB1

3DIARETDB1

DIARETDB1

DIARETDB0

CHASE-DB1

OC

Bin Rushed

Magrabi

标注相關工具

Tools	introduce	Reference	Introduce
labelme	分割 、分類	https://blog.csdn.net/heiheiya/article/details/88342597	最著名的标注工具之一，雖然其使用者界面有點慢，特别是縮放高清圖像時。
RectLabel	簡單易用，隻在 Mac 可用。
LabelBox	對于大型标記項目很合适，提供不同類型标記任務的選項。
COCO UI	用于标注 COCO 資料集的工具。

資料集相關代碼

dataset	-	introduce	github
Grishti-GS1	2015	An adaptive threshold based algorithm for optic disc and cup segmentation in fundus images	https://github.com/NupurBhaisare/Cup-and-disc-segmentation-for-glaucoma-detection-CDR-Calculation-
Grishti-GS1	MICCAI 2018	Towards a glaucoma risk index based on simulated hemodynamics from fundus images	https://github.com/ignaciorlando/glaucoma-hemodynamics
Grishti-GS1 RIMONE	keras	GAN For Glaucome segmentation	https://github.com/tomazrvb/Paper/tree/d638050a3cad67a9c00bd9977be8a01877bdc5e7
Grishti-GS1	2017	F21 複現	https://github.com/abhinav-iiit/fundus-image-segmentation
Grishti-GS1	2017 CVPR keras	F21	https://github.com/seva100/optic-nerve-cnn
-	pytorch	retinal-cGAN	https://github.com/shuangyueliao/retinal-cGAN

論文

常見會議&期刊

⭐	TOP	introduce
1	ISBI	The IEEE International Symposium on Biomedical Imaging	⭐
2	MICCA	International Conference on Medical Image Computing and Computer-Assisted Intervention	醫學影像分析 (Medical Image Analysis) 研究領域的頂尖年會
3	CVPR	IEEE Conference on Computer Vision and Pattern Recognition	計算機視覺和模式識别領域的頂級會議
4	IEEE TMI	transactions on medical imaging	醫學圖像處理頂級的頂級期刊,生物醫學圖像。
5	-	Medical Image Analysis	醫學影像分析期刊
6	MICS	國内醫學圖像領域規模最大的學術會議之一

論文期刊會議索引

N	-	-	introduce	-
4	IPMI	Information Processing in Medical Imaging	⭐
4	-	OPHTHALMOLOGY	SCI 1
5	KBS	Knowledge-Based Systems	SCI
6	-	Ophthalmology Glaucoma	SCI 2
7	ICTAI	IEEE International Conference on Tools with Artificial Intelligence
8	ICIP	IEEE International Conference on Image Processing
9	SSD
10	-	scientific reports	SCI 3
11	Information Sciences	SCI 2
12	PLOS ONE	SCI 3
-	PLOS MED	SCI 1
13	IJACSA	International Journal of Advanced Computer Science and Applications
14	T-MI	IEEE Transactions on Medical Imaging	SCI 2
15	EMBC	IEEE Engineering in Medicine and Biology Society
16	Computerized Medical Imaging and Graphics	SCI 4	36
17	Journal of Intelligent Systems	SCI 2
18	BMC Medical Imaging	SCI 4
-	BMC Ophthalmology	SCI 4
19	IOVS	investigatice ophthalmology & visual science	SCI 2
20	OMIA	Ophthalmic Medical Image Analysis	2/1
21	American Journal of Ophthalmology	SCI 2
22	Journal of Glaucoma	SCI 4
23	JAMA Ophthalmol	SCI 2	35

其他

N	-	-	introduce
PRCV	中國模式識别與計算機視覺大會
CCF-GAIR	全球人工智能與機器人峰會

方法彙總

Classification Segmentation Detection	Fundus OCT VF	No.	Position	Method	Introduce
C	F	1	global Image	Inception-v3	三分類(unlikely,suspect, certain)
C+S	F	2	global Image + domain knowledge feature	Faster Rcnn，CNN，FCN,MB-NN （multi-brance neural network model）
C	F	3	OD	difference-of-Gaussian blob detector + Resnet50	二分類(青光眼/正常眼)
S	F	4	OD/OC	定位 Daubechies wavelet transform + 去血管 + 9-layer CNN	CDR
S	F	5	OD/OC	FCN + Post-processing	CDR
S+C	F	6	OD/OC	ROI extraction + Multi-task CNN + Post-processing	二分類(青光眼/正常眼)
C	F	7	OD	CNN+SVM	二分類(青光眼/正常眼)
C	F	8	ONH	transfer Resnet50	二分類(青光眼/正常眼)評估了幾種深度學習架構和遷移學習
C	F	9	global Image	18 layer CNN	二分類(青光眼/正常眼)
S+C	F	10	OD	ROI+Extraction+DNN(SAE)+ASM	減弱PPA對分割OD影響
C	F	11	OD	CNN（3con+3maxpooling+2FC）	二分類(青光眼/正常眼)晚期和早期青光眼檢測
C	F	12	global Image	Glaucoma-Deep（CNN，DBM，softmax）	二分類(青光眼/正常眼)
C	F	13	global Image	ResNet	CNN、SVM、Random Forest對比
S	F	14	OD、OC	Polar coordinate + M-Net (Multi-Scale input layer + U-Net +Side-Output Layer + Multi-Label Loss Function)	CDR
C	F	15	global Image + local Image(optic disk region)	DENet（4個流）	Disc-aware,Multi-level ,multi-model
C	F	16	global Image	CNN (4con+4fc)	二分類(青光眼/正常眼)
C	F	17	global Image	基于RNN的更新檔分類來劃分真實的RNFLD邊界像素	RNFLD 邊界
S	F	18	global Image	Unet （encoder ：pre-trained ResNet34）	RIGA 訓練；DRISHTI-GS1、RIM-ONE測試
C	F	19	global Image	CNN+SVM	ROI deformable shape model 檢測 OD邊界， 內建了局部和整體特征
C	F	20	global Image	AG-CNN ( Attention 、Guide BP 、 ResNet)	二分類(青光眼/正常眼) 建立了一個注意力dataset 與F01/F16對比
S	F	21	OD/OC	crop ROI + FC-DenseNet （ FCN + DenseNet）+ Refinement
S	F	34	cGAN (generator + discriminator)	分割OD

結果彙總

Paper	Method	Dataset	OD/ DC(F)	OD/ JC(O)	OD / Acc	OC / DC(F)	OC /JC(O)	OC /Acc	SE	SP
21	Modified U-Net CNN (DL)	Drishti-GS	0.85	0.75	-	-	-	-
35	Large pixel patch based CNN (DL)	Drishti-GS	0.9373	0.8775	-	-	-	-
33	Ensemble learning based CNN (DL)	Drishti-GS	0.871	0.85	-	0.973	0.914	-
23	Fully convolutional DenseNet (DL)	Drishti-GS	0.8282	0.7113	0.9948	0.949	0.9042	0.9969
14	multi-label deep learning and Polar transformation (DL)	ORIGA	-	0.77	-	-	0.929	-
23	Fully convolutional DenseNet(DL)	ORIGA	0.8659	0.7688	-	0.9653	0.9334	-
21	Modified U-Net CNN (DL)	RIM-ONE	0.82	0.69	-	0.94	0.89	-
36	Fully convolutional and adversarial network (DL)	RIM-ONE	0.94	0.768	0.977	0.897
21	Modified U-Net CNN (DL)	DRIONS-DB	0.94	0.89
23	Fully convolutional DenseNet(DL)	DRIONS-DB	0.9415	0.8912
23	Fully convolutional DenseNet(DL)	ONHSD	0.9556	0.9155	0.999

名詞解析

青光眼 GON

GON	mean	chinese
CDR	cup-to disc ratio	杯盤比 杯區最大垂直高度除以盤區最大垂直高度
cpRNFLT	circumpapillary retinal nerve fiber layer thickness	視網膜乳頭周圍神經纖維層厚度
FOV	field of views	視野
GCS	ganglion cells	神經節細胞
GCC	ganglion cell complex	神經節細胞複合體
ISNT	inferior, superior, nasal and temporal
IOP	intraocular pressure	眼壓
mIRT	macular inner retinal thickness	黃斑内視網膜厚度
ONH	optic nerve head	視神經乳頭
OD	optic disc	視盤
OC	optic cup	視杯
ODD	optic disc diameter	視杯直徑
PPA	peripapillary atrophy	視盤旁萎縮弧
PPA	parapapillary atrophy	萎縮弧
POAG	primary open-angle glaucoma	原發性開角型青光眼
PACG	primary angle closure glaucoma	原發性閉角型青光眼
RNFLD	retinal nerve fiber layer defects	視網膜神經纖維層缺損
RPE	Retinal Pigment Epithelium	視網膜色素上皮細胞
RNFL	retinal nerve fiber layer	視網膜神經纖維層
RGC	retinal ganglion cells	視網膜神經節細胞
RGC+	retinal ganglion cell plus inner plexiform layer	視網膜神經節細胞加内叢狀層
VCD	vertical cup diameter	垂直杯徑
VDD	vertical disc diameter	垂直盤直徑
VF	visual fields
-	sclera	鞏膜
-	choroid	脈絡膜
-	macula	黃斑
-	vasculature hemorrhage	血管出血
-	cornea	角膜
-	neuroretinal rim	視網膜邊緣

儀器

image	mean	chinese
OCT	optical coherence tomography	光學相幹斷層成像術
CFI	color fundus imaging	彩色眼底成像
MRI	magnetic resonance imaging	磁共振成象
FOV	field of view	視野
CAD	Computer aided diagnosis	一種利用數字眼底圖像對青光眼做早期診斷的無創技術
GDx	scanning laser polarimetry (GDx: Carl Zeiss Meditec, Dublin, CA)	掃描雷射偏振測量儀
HRT	Heidelberg Retina Tomograph Heidelberg Engineering GmbH，Heidelberg，Germany	海德堡視網膜層析x射線攝影機
RS-3000 Advance	spectral-domain OCT (RS-3000 Advance, Nidek, Gamagori, Japan) images	光譜域OCT
OA-1000	OA-1000 optical biometer (Tomey, Nagoya, Japan)	光學生物儀 （測量軸向長度）
RC-5000	RC-5000 refract-keratometer (Tomey)	折射角膜曲率計（記錄屈光不正）
RC-5000	RC-5000 non-contact tonometer (Tomey)	非接觸眼壓計（測定IOP）
-	Goldmann applanation tonometer	壓平式眼壓計（測定IOP）
-	Kowa nonmyd WX camera	後眼底照片

其他疾病

Other disease	mean	chinese
DR	diabetic retinopathy	糖尿病視網膜病變
-	ophthalmology	眼科
-	cataract	白内障
AMD	age related macular degeneration	年齡相關性黃斑變性
BCVA	Best-corrected visual acuity	最佳矯正視力
logMAR	logarithm of the mimimum angle of resolution	最小分辨角的對數
ANOVA	one-way analysis of variance	單因素方差分析
-	myopia	近視

評價名額

Evalution	mean	chinese	Math
AUC	area under the receiver opterating characteristic curve	接收機光電特性曲線下面積
ROC	receiver operating characteristic	受試者工作特征
AROC	receiver operating characteristic curve	接受者操作特征曲線
IoU	intersection over union	交并比	I o U = T P T P + F P + F N IoU = \frac{ TP}{TP+FP+FN} IoU=TP+FP+FNTP​
SD	standard deviation	标準偏差
CI	confidence interval	置信區間
MD	mean deviation	平均偏差
FP	False Positive	假正例
FN	False Negative	假負例
TP	True Positive	真正例
TN	True Negative	真負例
AC	accuracy	正确率	A C = T P + T N T P + T N + F P + F N AC = \frac{ TP+TN }{TP+TN+FP+FN} AC=TP+TN+FP+FNTP+TN​
SE	sensitivity	靈敏度，将實際有病的人正确地判定為患者的比例。	S E = T P T P + F N SE=\frac {TP}{TP+FN} SE=TP+FNTP​
SP	specificity	特異度，将實際無病的人正确地判定為非患者的比例。	S P = T N T N + F P SP=\frac {TN}{TN+FP} SP=TN+FPTN​
PC	precision	精确率針對所有正例(TP+FP)而言 其中真正例(TP)占的比例 精确率也叫查準率	P C = T P T P + F P PC=\frac {TP}{TP+FP} PC=TP+FPTP​
PPV	positive predictive value	正例預測值
E	overlapping error	重疊的錯誤	E = 1 − A r e a ( S ∩ G ) A r e a ( S ∪ G ) E = 1 - \frac{Area(S\cap G) } {Area(S \cup G )} E=1−Area(S∪G)Area(S∩G)​
A	balanced accuracy	平均正确率
$ δ_E $	CDR error	CDR減去ground truth CDR值	$$δ_E=
-	Euclidean distance	歐幾裡得距離 (mean ± standard deviation)
-	Dice coefficient	$$Dice =2× \frac{
-	Jaccard Index	實質是 IoU	$$JS=\frac {

Example，10病人 7P，3N；

分類結果 正類6個（4P，2N） 負類4個（1N,3P）

TP : 4

TN : 2

FN : 1

FP : 3

I o U = T P T P + F P + F N IoU = \frac{ TP}{TP+FP+FN} IoU=TP+FP+FNTP​

E = 1 − A r e a ( S ∩ G ) A r e a ( S ∪ G ) E = 1 - \frac{Area(S\cap G) } {Area(S \cup G )} E=1−Area(S∪G)Area(S∩G)​

A = 1 2 （ S E + S P ） A= \frac{1}{2}（SE + SP） A=21​（SE+SP）

A C = T P + T N T P + T N + F P + F N AC = \frac{ TP+TN }{TP+TN+FP+FN} AC=TP+TN+FP+FNTP+TN​

S E = T P T P + F N SE=\frac {TP}{TP+FN} SE=TP+FNTP​

S P = T N T N + F P SP=\frac {TN}{TN+FP} SP=TN+FPTN​

P C = T P T P + F P PC=\frac {TP}{TP+FP} PC=TP+FPTP​

F 1 = 2 ∗ P C ∗ S E P C + S E F1=2 * \frac {PC*SE}{PC+SE} F1=2∗PC+SEPC∗SE​

F − s o c r e = 2 ∗ T P 2 ∗ T P + F P + F N F-socre=\frac{2*TP}{2*TP+FP+FN} F−socre=2∗TP+FP+FN2∗TP​

δ E = ∣ C D R S − C D R G ∣ δ_E = | CDR_S - CDR_G | δE​=∣CDRS​−CDRG​∣

D i c e = 2 × ∣ S ∩ G ∣ ∣ S ∣ + ∣ G ∣ Dice =2× \frac{|S\cap G|}{|S|+ |G|} Dice=2×∣S∣+∣G∣∣S∩G∣​

J S = ∣ G T ∩ S R ∣ ∣ G T ∪ S R ∣ = ∣ G T ∩ S R ∣ ∣ G T ∣ + ∣ S R ∣ − ∣ G T ∩ S R ∣ JS=\frac { |{GT}\cap{SR}| }{|GT \cup SR|}=\frac { |{GT}\cap{SR}| }{|GT |+|SR|-|{GT}\cap{SR}|} JS=∣GT∪SR∣∣GT∩SR∣​=∣GT∣+∣SR∣−∣GT∩SR∣∣GT∩SR∣​