Ukuthuthukisa Ukunakekelwa Kwesiguli Nge-AI-based Attenuation Correction in PET Imaging

Ucwaningo olusha olunesihloko esithi “Utilizing Pix-2-Pix GAN for Deep Learning-based Whole-Body PSMA PET/CT Attenuation Correction” lusanda kushicilelwa kuVolumu 15 ye-Oncotarget ngoMeyi 7, 2024.

Ukuchayeka emisebeni evela ezifundweni ezilandelanayo ze-PET/CT ekulandeleni isiguli se-oncology kuyakhathaza. Kulolu phenyo lwakamuva, ithimba labacwaningi abahlanganisa uKevin C. Ma, u-Esther Mena, uLiza Lindenberg, uNathan S. Lay, uPhillip Eclarinal, uDeborah E. Citrin, uPeter A. Pinto, uBradford J. Wood, uWilliam L. Dahut, uJames L. Gulley, uRavi A. Madan, uPeter L. Choyke, u-Ismail Baris Turkbey we-National Institute of Cancer Institute, u-Ismail Baris Turkbey kanye no-A. Izempilo yethule ithuluzi le-artificial intelligence (AI). Leli thuluzi lihlose ukukhiqiza izithombe ze-PET (AC-PET) ezilungiswe ukuncishiswa kusuka ezithombeni ze-PET (NAC-PET) ezingalungisiwe, okungenzeka zinciphise isidingo samaskena e-CT anedosi eliphansi.

"Izithombe ze-PET ezikhiqizwe nge-Ai zinamandla omtholampilo okunciphisa isidingo sokulungiswa kwe-CT scan ngenkathi kugcinwa izimpawu zobuningi kanye nekhwalithi yesithombe ezigulini ezinomdlavuza wendlala yesinye."

Izindlela: I-algorithm yokufunda ejulile esekelwe ku-2D Pix-2-Pix generative adversarial network (GAN) architecture yasungulwa ngokusekelwe ezithombeni ezibhanqiwe ze-AC-PET ne-NAC-PET. I-18F-DCFPyL PSMA (i-Prostate-specific membrane antigen) Ucwaningo lwe-PET-CT lweziguli ze-302 ezinomdlavuza we-prostate lwahlukaniswa lwaba ukuqeqeshwa, ukuqinisekiswa, namaqembu okuhlola (n 183, 60, kanye ne-59, ngokulandelana). Imodeli yaqeqeshwa kusetshenziswa amasu amabili amisiwe: I-Standard Uptake Value (SUV) esuselwe kanye ne-SUV-NYUL. Ukuskena ukusebenza okuvundlile kwahlaziywa kusetshenziswa iphutha le-normalized mean square square (NMSE), mean absolute error (MAE), structural similarity index (SSIM) kanye ne-peak signal-to-noise ratio (PSNR). Udokotela wezokwelapha nge-nuclear kungenzeka ukuthi wenze ukuhlaziya kwezinga lezilonda endaweni ethakaselwayo. Izinkomba ze-SUV zihlolwe kusetshenziswa i-intra-group coefficient (ICC), i-repeability coefficient (RC), kanye namamodeli wemiphumela exubile yomugqa.

Imiphumela:Eqenjini lokuhlola elizimele, i-NMSE emaphakathi, i-MAE, i-SSIM, ne-PSNR bekungu-13.26%, 3.59%, 0.891, kanye no-26.82, ngokulandelana. I-ICC ye-SUVmax ne-SUVmean ibingu-0.88 no-0.89, okubonisa ukuhlobana okuqinile phakathi komaka bezithombe zangempela nabakhiqizwe be-AI. Izinto ezifana nendawo yesilonda, ukuminyana (amayunithi e-Hounsfield), nokuthathwa kwesilonda kutholwe ukuthi kunomthelela ephutheni elihlobene kumamethrikhi e-SUV akhiqiziwe (konke p <0.05).

"I-AC-PET ekhiqizwe imodeli ye-Pix-2-Pix GAN ibonisa amamethrikhi e-SUV aqondana eduze nezithombe zangempela. Izithombe ze-PET ezikhiqizwe yi-AI zibonisa amandla athembisayo omtholampilo okunciphisa isidingo se-CT scan ukuze kulungiswe ukuthomba kuyilapho kugcinwa izimpawu zobuningi nekhwalithi yesithombe."

—————————————————————————————————————————————————————————————————————————————————————

Njengoba sonke sazi, ukuthuthukiswa kwemboni yezithombe zezokwelapha akunakuhlukaniswa nokuthuthukiswa kochungechunge lwemishini yezokwelapha - ama-ejenti e-agent ehlukile kanye nezinto ezisetshenziswayo ezisekelayo - ezisetshenziswa kakhulu kulo mkhakha. E-China, edume ngemboni yayo yokukhiqiza, kunabakhiqizi abaningi abadumile ekhaya naphesheya ngokukhiqizwa kwemishini yokuthwebula yezokwelapha, okuhlanganisa.I-LnkMed. Selokhu yasungulwa, i-LnkMed ibilokhu igxile emkhakheni wama-ejenti wokuqhathanisa nengcindezi ephezulu. Ithimba lonjiniyela be-LnkMed liholwa yi-Ph.D. unolwazi olungaphezu kweminyaka eyishumi futhi ubambe iqhaza ngokujulile ocwaningweni nasekuthuthukisweni. Ngaphansi kokuqondisa kwakhe, iCT umjovo wekhanda elilodwa,CT umjovo wekhanda kabili,Umjovo we-ejenti yokuqhathanisa ye-MRI, futhiUmjovo we-angiography wokucindezela okuphezulu kokungafaniaklanywe ngalezi zici: umzimba oqinile futhi ohlangene, isixhumi esibonakalayo esisebenzayo nesihlakaniphile, imisebenzi ephelele, ukuphepha okuphezulu, kanye nomklamo oqinile. Futhi singahlinzeka ngamasirinji namashubhu ahambisana nalezo zinhlobo ezidumile zamajovo e-CT,MRI,DSA Ngesimo sabo sengqondo esiqotho namandla ochwepheshe, bonke abasebenzi be-LnkMed bakumema ngobuqotho ukuthi nize nizohlola izimakethe ezengeziwe ndawonye.