@article{oai:hirosaki.repo.nii.ac.jp:00006138,
 author = {澤田, 洋平 and 澤谷, 学 and 三上, 達也 and 櫻庭, 裕丈 and 福田, 眞作},
 issue = {1},
 journal = {弘前医学},
 month = {Oct},
 note = {研究の目的は，自家蛍光内視鏡（AFI）を用いた組織型推論ニューラルネットワークシステムの試作，並びに精度検証である．
　AFI を用い，大腸病変と正常背景粘膜との自家蛍光強度を定量し，粘膜下層までの距離を反映する任意量（Intervalto submucosa 以下iTS）を求めた．更に，iTS の基本統計量，局在部位，肉眼型から組織型を推論するニューラルネットワークシステムを試作し精度評価をおこなった．対象はAFI で病変を観察し内視鏡的粘膜切除を施行した126病変（低異型度腺腫71，高異型度腺腫20，粘膜内癌25，鋸歯状腺腫／ポリープ10）である．病変は周辺正常粘膜よりもiTS が有意に高いことが分かった．粘膜内癌が鋸歯状腺腫／ポリープよりiTS が有意に高かったが，その他の組織型の間に有意差は認められなかった．一方，ニューラルネットワークシステムの層化2 分割交差検証による精度は80.2%であった．今後，汎化による実装が期待された.
 The aims of study are to develop a trial neural network system for automated classification of colorectalneoplasms by using autofluorescence imaging（ AFI）, and to evaluate system performance.
 By using autofluorescence imaging（ AFI）, comparative study on autofluorescence intensity that was quantified asinterval to submucosa（ iTS） among colonic tumors and its surrounding normal mucosa were conducted. In addition,neural network system for automated tumor classification was developed by attributes including the basic statisticalamount of iTS, tumor location or morphological type. A total of 126 AFI images （low grade adenoma71, high gradeadenoma20, mucosal cancer25, sessile serrated adenoma/polyp10） were studied. All the lesions had significantlyhigher iTS than its surrounding normal mucosa. Mucosal cancer presented with higher iTS when comparing withsessile serrated adenoma/polyp. No significant difference in iTS was demonstrated among other lesions. Accuracy ofthe trained neural network system evaluated by stratified 2-fold cross-validation was found to be 80.2%. For clinical use in automated tumor classification, further generalization of the system performance must be required.},
 pages = {65--70},
 title = {自家蛍光内視鏡による大腸腫瘍性病変の蛍光強度の定量分析とニューラルネットワークによる自動診断システムの試作},
 volume = {71},
 year = {2020}
}