Vi羽c hu医n luy羽n mơ h·nh R2U-Pgv"v逢挨pi"v詠 nj逢"swƒ tr·nh hu医n luy羽n mơ h·nh Vess-Pgv0"Ak吋m khƒc nhau gi英a hai quƒ tr·nh n y l quƒ tr·nh hu医n luy羽n mơ h·nh R2U-Net c„ tj‒o"d逢噂c ti隠n x穎 l#pj逢"j·nh 3.5
Hình 3.5: Quá trình hu医n luy羽n mơ hình R2U-Net
D逢噂c ti隠n x穎 l# n y th詠c hi羽n vi羽c 8逢c"違nh v ng m衣c m逸t v隠違nh xƒm v chu育n h„a 違nh trên ch nh 違nh xƒm n y ucw"8„"v<pi"8瓜"v逢挨pirj違p"d茨pi"ENCJG". Hình 3.6 o»"v違"d逢噂e"vk隠p"z穎"n#0
Hình 3.6<"D逢噂c ti隠n x穎 lý trong quá trình hu医n luy羽n mơ hình R2U-Net 3.2.3.Rj逢挨pi"rjƒr"rj¤p"nq衣i s穎 d映ng mơ hình SVM
Ucw"mjk"rj¤p"8q衣p"d茨pi"jck"rj逢挨pi"rjƒr"vt‒p違pj"u胤"8逢嬰e"8逢c"x q"o瓜v"UXO"8吋" rj¤p"nq衣k"zgo"違pj"e„"rj違k"違pj"d医v"vj逢運pi"jc{"mj»pi0"J·pj 3.7 o»"v違"sw{"vt·pjjw医p"
Hình 3.7: Quy trình hu医n luy羽n và phân lo衣i c栄a SVM
D英 li羽w"8逢嬰c s穎 d映ni"8吋 hu医n luy羽n mơ h·nh SVM l 違pj"rj¤p"8q衣n b茨ng tay c栄a cƒc t壱p HRF, DRIVE v STARE. T壱p d英 li羽u n y bao g欝m 50 違pj"ejkc"8隠u cho hai l噂p.
Cƒc d逢噂c c栄a quƒ tr·nh xƒe"8鵜nh b医v"vj逢運ng b茨ng SVM:
‚ D逢噂c 1: Chu育n h„a d英 li羽u: 謂pj"rj¤p"8q衣n s胤8逢嬰e"vjc{"8鰻i kej"vj逢噂c th nh 512x512.
‚ D逢噂c 2: Tr ch xu医v"8員e"vt逢pi<"O瓜v"xgevqt"e„"m ej"vj逢噂e"62726:"u胤"8衣k"fk羽p" ejq"eƒe"8員e"vt逢pi"vt ej"zw医v"8逢嬰e"v瑛"o瓜v"違pj0""Eƒe"8員e"vt逢pi"8逢嬰e"vt ej" zw医v"v瑛"違pj"n亥p"n逢嬰v"pj逢"ucw<"Ikƒ tr鵜8瓜 l噂n c栄a cƒe"rkzgn."8員e"vt逢pi"JQI." c医u tr¿c Haralick, mơ-men Hu.
‚ D逢噂c 3: Hu医n luy羽n: Mơ hình SVM cĩ kernel là tw{院p"v pj"*nkpgct+"u胤"8逢嬰e" jw医p"nw{羽p"d茨pi"eƒe"xgevqt"8員e"vt逢pi"vjw"8逢嬰e"荏"d逢噂e"4"v瑛"8„"e„"8逢嬰e"d瓜" rj¤p"nq衣k"UXO0
‚ D逢噂c 4: Ki吋m th穎: 謂pj"rj¤p"8q衣n v ng m衣c m逸t c„8逢嬰c t瑛 cƒe"rj逢挨pi" phƒp vt‒p"8k"swc"d瓜 phân lo衣k"UXO"8吋8逢嬰c phân lo衣i xem l c„ b医t vj逢運ng hay khơng.
3.3. Rj逢挨pi"rjƒr"8ƒpj"ikƒ
Jk羽w"uw医v"e栄c"rj逢挨pi"rjƒr"rj¤p"8q衣p"8逢嬰e"8ƒpj"ikƒ"d茨pi"eƒe"vj»pi"u嘘"x隠"8瓜"8q" pj逢"8瓜"pj衣{"*ugpukvkxkv{+."8瓜"8員e"jk羽w"*urgekhkekv{+"x "8瓜"ej pj"zƒe"*ceewttce{+0 Các vj»pi"u嘘"8逢嬰e"v pj"vqƒp"d茨pi"eƒej"uq"uƒpj"違pj"m院v"sw違"e栄c"swƒ"vt·pj"rj¤p"8q衣p"x噂k" 違pj"itqwpf"vtwvj"*違pj"8逢嬰e"rj¤p"8q衣p"vj栄"e»pi"d荏k"ejw{‒p"ikc+0
Vtqpi"8„"<
‚ A瓜"pj衣{"n "v益"n羽"eƒe"8k吋o"違pj"8逢嬰e"rj¤p"nq衣k"ej pj"zƒe"n "o衣ej0"M#"jk羽w" là SN.
‚ A瓜"8員e"jk羽w"n "v益"n羽"rjƒv"jk羽p"eƒe"8k吋o"違pj"mj»pi"n "o衣ej"ejnh xác. Ký jk羽w"n "UR0
‚ A瓜"ej pj"zƒe"n "v益"n羽"ik英c"u嘘"8k吋o"違pj"8逢嬰e"rj¤p"nq衣k"ej pj"zƒe"x噂k"u嘘" 8k吋o"違pj"vtqpi"違pj0"M#"jk羽w"n "CEE. E»pi"vj泳e"v pj"e栄c"eƒe"vj»pi"u嘘"vt‒p"pj逢"ucw"< 鯨軽 噺 "劇鶏 髪 繋軽劇鶏 鯨鶏 噺 "劇軽 髪 繋鶏劇軽 畦系系 噺 " 劇鶏 髪 劇軽 劇鶏 髪 繋軽 髪 劇軽 髪 繋鶏 Vtqpi"8„"< ‚ VR"n "u嘘"rkzgn"rj¤p"nq衣k"8¿pi"mjk"o瓜v"8k吋o"違pj"8逢嬰e"zƒe"8鵜pj"n "o衣ej" oƒw"vtqpi"e違"違pj"rj¤p"8q衣p"x "違pj"itqwpf"vtwvj0 ‚ VP"n "u嘘"rkzgn"rj¤p"nq衣k"8¿pi"mjk"o瓜v"8k吋o"違pj"8逢嬰e"zƒe"8鵜pj"n "p隠p"trong e違"違pj"rj¤p"8q衣p"x "違pj"itqwpf"vtwvj0 ‚ HR"n "u嘘"rkzgn"rj¤p"nq衣k"pj亥o"mjk"o瓜v"8k吋o"違pj"8逢嬰e"zƒe"8鵜pj"n "o衣ej" oƒw"vtqpi"違pj"rj¤p"8q衣p"pj逢pi"mj»pi"rj違k"n "o衣ej"vtqpi"違pj"itqwpf" vtwvj."jc{"e”p"n "rj¤p"nq衣k"pj亥o"vj pj"o衣ej0
‚ HP"n "u嘘"rkzgn"rj¤p"nq衣k"pj亥o"mjk"o瓜v"8k吋o"違pj"8逢嬰e"zƒe"8鵜pj"n "p隠p vtqpi"違pj"rj¤p"8q衣p"pj逢pi"n "o衣ej"vtqpi"違pj"itqwpf"vtwvj."jc{"e”p"i丑k" n "rj¤p"nq衣k"pj亥o"vj pj"p隠p.
EJ姶愛PI"6< HI烏N TH衛E"XÉ"AèPJ"IKè"M蔭T QU謂
Vtqpi"ej逢挨pi"p {."vƒe"ik違"vt·pj"d {"{‒w"e亥w"rj亥p"e泳pi."m院v"sw違"jk羽p"vj詠e"e栄c" ik違k"vjw壱v"8隠"zw医v0
4.1. Yêu c亥u ph亥n c泳ng
Mơ h·nh Vess-Net, R2U-Net v SXO"8逢嬰c train trên Google Colab pro c„ c医u h·pj"pj逢"ucw<
‚ H羽8k隠u h nh: Ubuntu 18.04.5 LTS.
‚ B瓜 x穎 l#: Intel(R) Xeon(R) CPU @ 2.20GHz. ‚ GPU: NVIDIA Tesla P100.
‚ Xung nh鵜p: 2200 GHz.
‚ B瓜 nh噂 Ram: 25GB.
4.2. T壱p d英 li羽u s穎 d映ng
A瓜"jk羽w"sw違"e栄c"ik違k"vjw壱v"Xguu-Pgv"8逢嬰e"8ƒpj"ikƒ"vt‒p"d嘘p"v壱r"f英"nk羽w"n "FTKXG. STARE, CHASE_DB1, JTH0"A瓜"jk羽w"sw違"e栄c"ik違k"vjw壱v"T4W-PGV"8逢嬰e"8ƒpj"ikƒ" vt‒p"jck"v壱r"f英"nk羽w"n "UVCTG."EJCUGaFD30
Cƒc t壱p d英 li羽u n y ch泳a cƒc 違nh v ng m衣c m逸v"8逢嬰c thu th壱p 荏 cƒe"8k隠u ki羽n khƒc nhau v 違pj"8逢嬰e"rj¤p"8q衣n b茨ng tay c栄a n„ 8逢嬰c th詠c hi羽n b荏i cƒc chuyên gia (違nh ground truth). K cj"vj逢噂c 違nh v s嘘n逢嬰ng 違nh c栄a m厩i t壱p d英 li羽w"8逢嬰c mơ t違 trong b違ng 4.1. H·nh 4.1 minh h丑a 違nh m磯w"8逢嬰c l医y trong cƒc t壱p d英 li羽u trên.
T壱p d英 li羽u M ej"vj逢噂c S嘘n逢嬰ng 違nh DRIVE 565 x 584 40 CHASE_DB1 999 x 960 28 STARE 700 x 605 20 HRF 3504 x 2336 45 B違ng 4.1: Thơng s嘘 c栄a các t壱p d英 li羽u 4.3. K院t qu違 th詠c hi羽n
4.3.1.Rj逢挨pi"rjƒr"rj¤p"8q衣n 違nh v ng m衣c m逸t s穎 d映ng m衣ng h丑c sâu Vess- Net
M院v"sw違"rj¤p"8q衣p"u穎"f映pi"o衣pi"e„"nw欝pi"f逢"mfir"*Xguu-Pgv+"vt‒p"eƒe"v壱r"f英"nk羽w" 8逢嬰e"vj吋"jk羽p"荏"d違pi"602
Tên b瓜 d英 li羽u 謂pj"8亥u vào 謂nh ground truth 謂pj"8逢嬰c phân 8q衣n DRIVE
STARE
CHASE_DB1
4.3.2.Rj逢挨pi"rjƒr"rj¤p"8q衣n 違nh v ng m衣c m逸t s穎 d映ng m衣ng h丑c sâu R2U- Net
M院v"sw違"rj¤p"8q衣p"u穎"f映pi"o衣pi"T4W-Pgv"vt‒p"eƒe"v壱r"f英"nk羽w"8逢嬰e"vj吋"jk羽p"荏" d違pi"603.
Tên b瓜 d英 li羽u 謂pj"8亥u vào 謂nh ground truth 謂pj"8逢嬰c phân 8q衣n STARE
CHASE_DB1
B違ng 4.3: K院t qu違rj¤p"8q衣n s穎 d映ng m衣ng R2U-Net trên các t壱p d英 li羽u 4.4. Aƒpj"ikƒ"m院t qu違
4.4.1.Aƒpj"ikƒ"m院t qu違 trên t壱p DRIVE 4.4.1.1. Rj逢挨pi"rjƒr"u穎 d映ng m衣ng Vess-Net
M院v"sw違"8ƒpj"ikƒ"vt‒p"42"違pj"vguv"e栄c"v壱r"f英"nk羽w"FTKXG"f́pi"o衣pi"Xguu-Net 8逢嬰e"vj吋"jk羽p"荏"d違pi"604.
Hình 違nh SN SP ACC Phân l噂p 1 0.8413 0.9745 0.9593 Normal 2 0.7743 0.9844 0.9576 Normal 3 0.5258 0.9948 0.9364 Abnormal 4 0.7569 0.9859 0.9593 Normal 5 0.6983 0.9897 0.9549 Abnormal 6 0.6960 0.9884 0.9521 Abnormal 7 0.7223 0.9800 0.9501 Normal 8 0.6856 0.9861 0.9533 Abnormal 9 0. 6652 0.9912 0.9569 Abnormal 10 0.7638 0.9829 0.9597 Normal 11 0.7698 0.9712 0.9477 Normal 12 0.7789 0.9819 0.9597 Abnormal 13 0.7299 0.9852 0.9531 Abnormal 14 0.8153 0.9757 0.9592 Abnormal 15 0.7793 0.9767 0.9584 Normal 16 0.7874 0.9848 0.9622 Normal 17 0.6934 0.9877 0.9564 Abnormal 18 0.8074 0.9804 0.9630 Abnormal 19 0.8852 0.9773 0.9675 Normal 20 0.8197 0.9805 0.9655 Normal Trung bình 0.7498 0.9830 0.9567 0.7
B違ng 4.4: K院t qu違8ƒpj"ikƒ"vt‒p"v壱p DRIVE dùng m衣ng Vess-Net
T瑛 b違ng 4.4 cho th医y k院t qu違 thu 8逢嬰c trên t壱p DRIVE c栄a Vess-Net c„ thơng s嘘 ACC, SP khƒ cao (0.9567 v 0.9830) c”n ch雨 s嘘 SN vj医rj挨p"mjƒ nhi隠u so v噂i hai thơng s嘘 trên (0.7498). K院t qu違 vi羽c phân lo衣i 違nh b医t tj逢運ng c ng 荏 m泳c khƒ l 70% phân lo衣k"8¿ng.
4.4.2.Aƒpj giá k院t qu違 trên t壱p STARE 4.4.2.1. Rj逢挨pi"rjƒr"u穎 d映ng m衣ng Vess-Net
M院v"sw違"8ƒpj"ikƒ"vt‒p"42"違pj"vguv"e栄c"v壱r"f英"nk羽w"UVCTG"f́pi"o衣pi"Xguu-Net 8逢嬰e"vj吋"jk羽p"荏"d違pi"6070
Hình 違nh SN SP ACC Phân l噂p 1 0.6481 0.9951 0.9622 Abnormal 2 0.7785 0.9961 0.9784 Abnormal 3 0.9023 0.9950 0.9882 Abnormal 4 0.5520 0.9981 0.9597 Abnormal 5 0.8362 0.9494 0.9375 Normal 6 07438 0.9843 0.9642 Abnormal 7 0.9627 0.9714 0.9706 Normal 8 0.9620 0.9791 0.9775 Normal 9 0.9416 0.9838 0.9797 Normal 10 0.9295 0.9785 0.9737 Abnormal 11 0.8934 0.9827 0.9744 Normal 12 0.9497 0.9833 0.9802 Normal 13 0.9201 0.9858 0.9788 Normal 14 0.9307 0.9886 0.9795 Abnormal 15 0.7961 0.9905 0.9702 Abnormal 16 0.7286 0.9901 0.9593 Normal 17 0.8452 0.9784 0.9636 Normal 18 0.7192 0.9953 0.9785 Normal 19 0.5498 0.9976 0.9740 Normal 20 0.4887 0.9977 0.9582 Abnormal Trung bình 0.8026 0.9861 0.9705 0.75
B違ng 4.5: K院t qu違8ƒpj"ikƒ"vt‒p"v壱p STARE dùng m衣ng Vess-Net
T瑛 b違ni"607"8逢嬰c cho th医y, cƒc ch雨 s嘘CEE."UP"vjw"8逢嬰c trên t壱p STARE d́ng m衣ng Vess-Ngv"ecq"j挨p"uq"x噂i k院t qu違vjw"8逢嬰c trên t壱p DRIVE (0.9705 v噂i 0.9567, 0.8026 v噂i 0.7498). V c„ ch雨 s嘘 SP v k院t qu違 phân lo衣k"v逢挨pi"8逢挨pi"pjcw"*0.9861 v噂i 0.9830, 0.7 v噂i 0.75).
4.4.2.2. Rj逢挨pi"rjƒr"u穎 d映ng m衣ng R2U-Net
M院v"sw違"8ƒpj"ikƒ"vt‒p"42"違pj"vguv"e栄c"v壱r"f英"nk羽w"UVCTG"f́pi"o衣pi"T4W-Net 8逢嬰e"vj吋"jk羽p"荏"d違pi"6080
Hình 違nh SN SP ACC Phân l噂p 1 0.6540 0.8254 0.8073 Abnormal 2 0.5436 0.8212 0.7961 Abnormal 3 0.7607 0.8216 0.8167 Abnormal 4 0.3990 0.8586 0.8156 Nomal 5 0.7405 0.8197 0.8106 Nomal 6 0.8313 0.8662 0.8629 Abnormal 7 0.8868 0.8344 0.8401 Normal 8 0.8408 0.8343 0.8349 Normal 9 0.8172 0.8546 0.8506 Normal 10 0.7911 0.8137 0.8112 Abnormal 11 0.7759 0.8444 0.8370 Normal 12 0.8480 0.8583 0.8572 Normal 13 0.7844 0.8503 0.8425 Normal 14 0.7624 0.8471 0.8371 Abnormal 15 0.6984 0.8485 0.8310 Normal 16 0.6308 0.8736 0.8424 Normal 17 0.7677 0.8401 0.8424 Normal 18 0.7677 0.8401 0.8310 Normal 19 0.5156 0.8769 0.8555 Normal 20 0.4165 0.8717 0.8334 Abnormal Trung bình 0.7040 0.8461 0.8329 0.75
B違ng 4.6: K院t qu違8ƒpj"ikƒ"vt‒p"v壱p STARE dùng m衣ng R2U-Net
T瑛 b違ng 4.5 v 4.6 cho th医y k院t qu違 tjw"8逢嬰c trên t壱p STARE s穎 d映pi"jck"rj逢挨pi" c„ s詠 khƒc nhau r医t l噂n. C映 th吋 cƒc ch雨 s嘘 khi s穎 d映ng m衣ng R2U-Net th医r"j挨p"uq" v噂i s穎 d映ng m衣ng Vess-Net (ACC: 0.8329 so v噂i 0.9705, SP: 0.8461 so v噂i 0.9861, SN: 0.7040 so v噂i 0.8026 ). k院t qu違rj¤p"nq衣k"épi"n "2097.
4.4.3.Aƒpj"ikƒ"m院t qu違 trên t壱p CHASE_DB1 4.4.3.1. Rj逢挨pi"rjƒr"u穎 d映ng m衣ng Vess-Net
M院v"sw違"8ƒpj"ikƒ"vt‒p"42"違pj"vguv"e栄c"v壱r"f英"nk羽w"EJCUG_FD3"f́pi"o衣pi"Xguu- Pgv"8逢嬰e"vj吋"jk羽p"荏"d違pi"6090
B違ng 4.7 cho th医y k院t qu違vjw"8逢嬰c trên t壱p CHASE_DB1 d́ng m衣ng Vess-Net l cao nh医t so v噂i k院t qu違vjw"8逢嬰c t瑛 nh英ng t壱p d英 li羽u khƒc s穎 d映ng ćpi"rj逢挨pi" pháp m衣ng h丑c sâu Vess-Net c pi"pj逢"ecq"j挨p so v噂i k院t qu違8ƒnh giƒ trên cƒc t壱p d英 li羽u s穎 d映ng m衣ng h丑c sâu R2U-Net. C映 th吋 k院t qu違 n y l : ACC: 0.9835, SP: 0.9931, SN: 0.87910"Fq"vt‒p"v壱r"f英"nk羽w"EJCUGaFD3"mj»pi"e„"pj«p"rj¤p"nq衣k"違pj" p‒p"mj»pi"vj吋"8ƒpj"ikƒ"m院v"sw違"rj¤p"nq衣k"vt‒p"v壱r"p {0 Hình 違nh SN SP ACC Phân l噂p 1 0.9453 0.9979 0.9935 Normal 2 0.9007 0.9971 0.9885 Normal 3 0.8966 0.9962 0.9865 Normal 4 0.9104 0.9945 0.9866 Normal 5 0.9146 0.9965 0.9888 Normal 6 0.9360 0.9971 0.9916 Normal 7 0.8825 0.9969 0.9865 Normal 8 0.8514 0.9961 0.9831 Normal 9 0.8471 0.9940 0.9803 Normal 10 0.8387 0.9938 0.9783 Normal 11 0.8777 0.9975 0.9867 Normal 12 0.8951 0.9933 0.9845 Normal 13 0.9006 0.9974 0.9884 Normal 14 0.8806 0.9967 0.9862 Abnormal 15 0.8660 0.9857 0.9764 Normal 16 0.8788 0.9842 0.9762 Normal 17 0.7950 0.9876 0.9758 Normal 18 0.7484 0.9895 0.9746 Normal 19 0.9031 0.9864 0.9800 Normal 20 0.9133 0.9827 0.9773 Normal Trung bình 0.8791 0.9931 0.9835 -
B違ng 4.7: K院t qu違8ƒpj"ikƒ"vt‒p"v壱p CHASE_DB1 dùng m衣ng Vess-Net 4.4.3.2. Rj逢挨pi"rjƒr"u穎 d映ng m衣ng R2U-Net
M院v"sw違"8ƒpj"ikƒ"vt‒p"42"違pj"vguv"e栄c"v壱r"f英"nk羽w"EJCUG_FD3"f́pi"o衣pi"T4W- Pgv"8逢嬰e"vj吋"jk羽p"荏"d違pi"60:0
Hình 違nh SN SP ACC Phân l噂p 1 0.5280 0.9482 0.9093 Normal 2 0.4040 0.9724 0.9161 Normal 3 0.4181 0.9506 0.8928 Normal 4 0.4591 0.9620 0.9091 Abnormal 5 0.5178 0.9504 0.9044 Abnormal 6 0.4993 0.9603 0.9143 Normal 7 0.3736 0.9540 0.8961 Abnormal 8 0.4187 0.9542 0.9010 Normal 9 0.4587 0.9503 0.8999 Normal 10 0.4710 0.9487 0.8957 Normal 11 0.4206 0.9663 0.9119 Normal 12 0.4290 0.9645 0.9106 Normal 13 0.3574 0.9562 0.8938 Abnormal 14 0.3859 0.9522 0.8950 Normal 15 0.4926 0.9494 0.9095 Abnormal 16 0.5469 0.9491 0.9150 Normal 17 0.3377 0.9688 0.9253 Normal 18 0.3092 0.9593 0.9135 Normal 19 0.5251 0.9376 0.9016 Normal 20 0.5656 0.9543 0.9200 Normal Trung bình 0.4460 0.9555 0.9068 -
D違pi"4.8: M院v"sw違"8ƒpj"ikƒ"vt‒p"v壱r"CHASE_DB1 f́pi"o衣pi"R2U-Net
B違ng 4.7 v 4.8 cho th医y k院t qu違 tjw"8逢嬰c vt‒p"EJCUGaFD3"f́pi"o衣pi"T4W- Net th医p so v噂i k院t qu違vjw"8逢嬰c vt‒p"EJCUGaFD3"f́pi"o衣pi"Xgus-Pgv"8員c bi羽t l ch雨 s嘘 SN. C映 th吋 so sƒnh c栄a hai b違pi"vt‒p"pj逢"ucw"<"CEE<"0.9068 so v噂i 0.9835, SP: 0.9555 so v噂i 0.9931, SN: 0.4460 so v噂i 0.8791.
4.4.4.Aƒpj"ikƒ"m院t qu違 trên t壱p HRF 4.4.4.1. Rj逢挨pi"rjƒr"u穎 d映ng m衣ng Vess-Net M院v"sw違"8ƒpj"ikƒ"vt‒p"42"違pj"vguv"e栄c"v壱r"f英"nk羽w"JTH"f́pi"o衣pi"Xguu-Pgv"8逢嬰e" vj吋"jk羽p"荏"d違pi"60;0 Hình 違nh SN SP ACC Phân l噂p 1 0.7519 0.9779 0.9564 Abnormal 2 0.7540 0.9804 0.9566 Abnormal 3 0.6231 0.9917 0.9619 Abnormal 4 0.6817 0.9880 0.9602 Abnormal 5 0.7482 0.9849 0.9633 Abnormal 6 0.7570 0.9831 0.9616 Abnormal 7 0.7343 0.9826 0.9594 Abnormal 8 0.7486 0.9811 0.9583 Abnormal 9 0.7495 0.9815 0.9592 Abnormal 10 0.7412 0.9825 0.9594 Abnormal 11 0.7865 0.9763 0.9540 Normal 12 0.8236 0.9836 0.9644 Normal 13 0.5336 0.9892 0.9446 Abnormal 14 0.8153 0.9853 0.9627 Normal 15 0.5785 0.9890 0.9497 Normal 16 0.8146 0.9817 0.9629 Normal 17 0.6445 0.9818 0.9453 Normal 18 0.7905 0.9824 0.9606 Normal 19 0.7397 0.9702 0.9494 Abnormal 20 0.7551 0.9911 0.9673 Normal Trung bình 0.7286 0.9833 0.9579 0.85
D違pi"4.9: M院v"sw違"8ƒpj"ikƒ"vt‒p"v壱r"HRF f́pi"o衣pi"Xguu-Net
B違ng 4.9 cho th医y k院t qu違 rj¤p"8q衣n trên t壱p HRF khƒ cao (ACC: 0.9579, SP: 0.9833 v SN: 0.7286). K院t qu違 phân lo衣i 違nh trên t壱p HRF l 85% phân lo衣i 違nh ch nh xƒc.
4.4.5.So sánh k院t qu違eƒe"rj逢挨pi"rjƒr T壱p d英 li羽u Rj逢挨pi"rjƒr SN SP ACC STARE Vess-Net 0.8026 0.9861 0.9705 R2U-Net 0.7040 0.8461 0.8329 CHASE_DB1 Vess-Net 0.8791 0.9931 0.9835 R2U-Net 0.4460 0.9555 0.9068 D違pi"4.10: M院v"sw違"8ƒpj"ikƒ"uq"uƒpj"v鰻pi"swƒv"vt‒p"eƒe"v壱r"f英"nk羽w"mjƒe"pjcw V瑛"d違pi"6032"vc"vj医{"8逢嬰e"o衣pi"j丑e"u¤w"Xguu-Pgv"8«"8逢嬰e"jk羽p"vj詠e"vtqpi"8隠"v k" p {"e„"8瓜"jk羽w"sw違"ecq"j挨p"t "t羽v"uq"x噂k"o衣pi"j丑e"u¤w"T4W-Pgv"8«"8逢嬰e"jk羽p"vj詠e0"
A吋 8衣v"8逢嬰e"8瓜 hi羽u qu違 trên, Vess-Pgv"8« 8逢嬰c thi院t k院 8吋 bi吋u di宇n cƒe"8員c tr逢pi"e栄a 違nh v ng m衣c m逸t b茨ng cƒc feature map v瑛a 8栄 (32x32 v噂i 違nh c„ k ch vj逢噂c 734z734+"8吋 bi吋u di宇n chi ti院t cƒc m衣ch mƒu nh臼. Cƒe"8逢運ng d磯n ph亥p"f逢"p瓜i gi¿p trƒnh vi羽c m医t mƒt thơng tin trong quƒ tr·nh tr ch xu医v"8員e"vt逢pi"ik¿p vi羽c tr ch xu医t 8員c vt逢pi"p y 8衣t hi羽u qu違ecq"j挨p0"A逢運ng d磯n ph亥p"f逢"pgo衣i cung c医p thơng tin m瓜t cƒch tr詠c ti院p t瑛 b瓜 gpeqfgt"8院n b瓜 decoder gi¿p vi羽c ph映c j欝k 違nh c„ k院t qu違 t嘘v"j挨p0
V隠 mơ h·nh R2U-Net c„ k院t qu違 th医p cĩ nguyên nj¤p"荏"d逢噂c ti隠n x穎 l#, fq"ej逢c" 8逢c"x́pi"piq k"HQX"e栄c"違pj"x隠"ikƒ"vt鵜"2"p‒p"i¤{"pjk宇w"mj»pi"e亥p"vjk院v"vt‒p"違pj0 Xk羽e"vk隠p"z穎"n#"ej逢c"jk羽w"sw違"vt‒p"v壱r"EJCUGaFD3"f磯p"8院p"m院v"sw違"rj¤p"8q衣p"違pj" ej逢c"ecq0 Hình 4.2<"Uq"uƒpj"m院v"sw違"e栄c"jck"o»"j·pj"vt‒p"違pj"e栄c"v壱r"UVCTG0"(a): 違nh rj¤p"8q衣p"d荏k"ejw{‒p"ikc, (b): k院t qu違mjk"rj¤p"8q衣n b茨ng Vess-Net, (c): k院t qu違mjk"rj¤p"8q衣n b茨ng R2U-Net
EJ姶愛PI"7< K蔭T LU一N
Vtqpi"ej逢挨pi"p {."vƒe"ik違vt·pj"d {"m院v"sw違"8衣v"8逢嬰e."逢w"pj逢嬰e"8k吋o"e栄c"rj逢挨pi" rjƒr"8隠"zw医v"x "j逢噂pi"o荏"t瓜pi"vtqpi"v逢挨pi"nck0
5.1. K院t qu違 8衣v"8逢嬰c
雲"8隠"v k"p {."tƒe"ik違"8«"v·o"jk吋w và 8隠"zw医v jck"rj逢挨pi"rjƒr"rj¤p"8q衣p"違pj"x pi" o衣e"o逸v."f詠c"vt‒p"違pj"rj¤p"8q衣p"8吋"rjƒv"jk羽p"違pj"e„d医v"vj逢運pi0"Jk羽p"vj詠e"x "8ƒpj" ikƒ"8瓜"jk羽w"sw違"e栄c"rj逢挨pi"rjƒr"8隠"zw医v0
5.2. 姶w và pj逢嬰e"8k吋m c栄c"rj逢挨pi"rjƒr"8隠 xu医t
姶w"8k吋o."rj逢挨pi"rjƒr"u穎"f映pi"o衣pi"j丑e"u¤w"Xguu-Pgv"e„"8瓜"jk羽w"sw違"ecq, cho m院v"sw違"rj¤p"8q衣p"o衣ej"oƒw"t "pfiv."mj»pi"ej泳c"pjk宇w0
Pj逢嬰e"8k吋o. v噂keƒej"jk羽p"vj詠evtqpi"8隠"v k p {"o衣pij丑e"u¤w"T4W-Pgv"e„"m院v" sw違"vj医r"x ej泳c"pjk宇w"vt‒p"pj英pi"違pj"rj¤p"8q衣p"vjw"8逢嬰e. Swƒ"vt·pj"jw医p"nw{羽p" eƒe"o衣pi"j丑e"u¤w"v嘘p"pjk隠w"v k"piw{‒p0
5.3. J逢噂ng m荏 r瓜pi"v逢挨pi"nck
V·o"jk吋w"x "pijk‒p"e泳w"vj‒o"o瓜v"u嘘"ik違k"vjw壱v"rjƒv"jk羽pd医v"vj逢運pi (8„o"uƒpi). E違k"vk院p"rj逢挨pi"rjƒr"jk羽p"vj詠e"8吋"v<pi"8瓜"jk羽w"sw違"ejq"o衣pi"j丑e"u¤w"T4W-Net. Vj‒o"eƒe"d逢噂e"vk隠p"z穎"n#."j壱w"z穎"n#"8吋p¤pi"ecq"jk羽w"uw医ve栄c"o衣pi"j丑e"u¤w"Xguu- Net.
TÀI LI烏U THAM KH謂O
[1] Molina-Casado, José and Carmona, Enrique and García-Feijoĩ, Julián, "Fast detection of the main anatomical structures in digital retinal images based on intra- and inter-structure relational knowledge," Computer Methods and Programs in Biomedicine, vol. 149, pp. 55-68, 2017.
[2] https://vi.wikipedia.org/wiki/M%C3%B4_h%C3%ACnh_m%C3%A0u_RG B?fbclid=IwAR0eEovoS2g4XjkayMIBvBLJylFYXILRwvKLvJJ1QvR_3t6 2VhfiAQusWM, (vtw{"e壱r"n亥p"ew嘘k"36i52."pi {"30/12/2021).
[3] Marwan D. Saleh, C. Eswaran, Ahmed Mueen."ÐCp"Cwvqocvgf"Dnqqf"Xguugn" Segmentation Algorithm Using Histogram Equalization and Automatic Vjtgujqnf"UgngevkqpĐ."Journal of Digital Imaging, 2011.
[4] URL: https://drive.grand-challenge.org/, (truy c壱p l亥n cu嘘i 14g30,ngày 10/12/2021).
[5] https://itnavi.com.vn/blog/cnn-la-gi/, (truy c壱p l亥n cu嘘i 10g30,ngày 08/12/2021).
[6] https://topbinhduong.net/neural-network-la-gi-cung-nhu-convolutional- neural-network-la-gi/, (truy c壱p l亥n cu嘘i 10g30,ngày 08/12/2021). [7] https://www.pyimagesearch.com/2021/05/14/convolutional-neural-
networks-cnns-and-layer-types/, (truy c壱p l亥n cu嘘i 10g30,ngày 08/12/2021). [8] https://topdev.vn/blog/thuat-toan-cnn-convolutional-neural-network/, (truy
c壱p l亥n cu嘘i 10g30,ngày 08/12/2021).
[9] https://www.superdatascience.com/blogs/convolutional-neural-networks- cnn-step-4-full-connection, (truy c壱p l亥n cu嘘i 10g30,ngày 08/12/2021). [10] https://viblo.asia/p/gioi-thieu-mang-resnet-vyDZOa7R5wj, (truy c壱p l亥n cu嘘i
10g30,ngày 08/12/2021).
[11] https://ongxuanhong.wordpress.com/2015/09/19/support-vector-machine- svm-hoi-gi-dap-nay/, (truy c壱p l亥n cu嘘i 10g30,ngày 08/12/2021).
[12] https://towardsdatascience.com/transposed-convolution-demystified- 84ca81b4baba, (truy c壱p l亥n cu嘘i: 14g30 ng y 6/12/2021).
[13] https://www.researchgate.net/figure/An-illustration-of-the-switch-and- unpooling-operation-in-a-deconvolutional-network-Using_fig1_338263538, (truy c壱p l亥n cu嘘i 15g ng y 06/12/2021).
[14] Muhammad Arsalan, Muhammad Owais, Tahir Mahmood, Se Woon Cho, and Kang Ryoung Park."ÐCkfkpi"vjg"Fkcipquku"qh"Fkcdgvke"cpf"J{rgtvgpukxg" Retinopathy Using Artificial Intelligence-Dcugf" Ugocpvke" UgiogpvcvkqpĐ." Journal of Clinical Medicine(2019), pp. 1-10, 2019.
[15] Zhexin Jiang, Hao Zhang, Yi Wang, Seok-Bum Ko."ÐTgvkpcn"dnqqf"xguugn" ugiogpvcvkqp" wukpi" hwnn{" eqpxqnwvkqpcn" pgvyqtm" ykvj" vtcpuhgt" ngctpkpiĐ." Computerized Medical Imaging and Graphics(2018), Volume 68, pp. 1-15, 2018.
[16] Ana Salazar-Gonzalez, Djibril Kaba, Yongmin Li, Xiaohui Liu, ÐUgiogpvcvkqp"qh"vjg"Dnqqf"Xguugnu"cpf"Qrvke"Fkum"kp"Tgvkpcn"KociguĐ."IEEE Journal of Biomedical and Health Informatics(2014), Volume 18, pp. 1874- 1886, 2014.
[17] Sanyukta Chetia, S. R. Nirmala." ÐTgvkpcn" Dnqqf" Xguugn" Vqtvwqukv{" Ogcuwtgogpv"hqt"Cpcn{uku"qh"J{rgtvgpukxg"Tgvkpqrcvj{Đ."2017 International Conference on Innovations in Electronics, Signal Processing and Communication (IESC), 2017.
[18] Seung Yeon Shin, Soochahn Lee, Il Dong Yun, Kyoung Mu Lee." ÐFggr" Vessel Segmentation By Learning Graphical Cqppgevkxkv{Đ."Ogfkecn"Kocig" Analysis(2019) , 101556 , 2019.
[19] T. Jemima Jebaseeli, C. Anand Deva Durai, J. Dinesh Peter."ÐTgvkpcn"Dnqqf" Xguugn" Ugiogpvcvkqp" htqo" Fgrkiogpvgf" Fkcdgvke" Tgvkpqrcvj{" KociguĐ. Computers & Electrical Engineering, pp. 245-258, 2018.
[20] Md Zahangir Alom, Mahmudul Hasan, Chris Yakopcic, Tarek M. Taha, Vijayan K. Asari."ÐTgewttgpv"Tgukfwcn"Eqpxqnwvkqpcn"Pgwtcn"Pgvyqtm" based on U-Net (R2U-Pgv+"hqt"Ogfkecn"Kocig"UgiogpvcvkqpĐ."Rtqlgev<"Deep Learning for Big Data analytics on High Performance Computing System, 2018. [21] https://drive.grand-challenge.org, (truy c壱p l亥n cu嘘i 15g ng y 08/12/2021). [22] https://cecas.clemson.edu/~ahoover/stare/, (truy c壱p l亥n cu嘘i 15g ng y 08/12/2021). [23] https://blogs.kingston.ac.uk/retinal/chasedb1/?fbclid=IwAR0Acr0AcGpxVj ob9QRCx4CUffzmhPPz6qi0k5Ro9TxAdWUwzzeQ5ghg7I, (truy c壱p l亥n
[24] https://www5.cs.fau.de/research/data/fundus- images/?fbclid=IwAR1MpB4h9-
jD1H3JVrpUWnZ6Jc1fwOn4Ocf7lwG0GzOtZqT0d8vMUfY46KQ, (truy c壱p l亥n cu嘘i 15g ng y 08/12/2021).