演題抄録

ポスター

開催概要
開催回
第58回・2020年・京都
 

深層学習による早期子宮体癌における再発予測

演題番号 : P-244

[筆頭演者]
赤澤 宗俊:1 
[共同演者]
橋本 和法:1、須知 慧子:1、野口 乙:1、春日 みさき:1、岡本 英恵:1、古川 由理:1、立花 康成:1、上野 麻理子:1、一戸 晶元:1、長野 浩明:1

1:東京女子医科大学東医療センター・産婦人科

 

Purpose: Most women with early-stage endometrial cancer have a favorable prognosis. However, the 10 - 20 % of the patients develop recurrence after initial treatment of the primary. Mainly, gynecologist relied on FIGO stage for oncologic outcome, patients' background and therapeutic factors also affected the provability of the recurrence of early stage cancer. We tried to predict recurrence in early-stage endometrial cancer using deep learning methods based on patients' backgrounds, results of pathology and contents of surgeries and adjuvant therapy.

Methods: We enrolled 74 cases of early-stage endometrial cancers (FIGO stage I or II), who have received primary surgical treatment in our institute and follow-up until recurrence or 5-year after surgeries. Using deep learning classifiers consisting of 4 layers (1 input layer/ 2 hidden layers/ 1 output layer), we predicted recurrence from 16 parameters, including patients' demographics (age, BMI, gravity/parity, hypertension/diabetic), pathologic factors (stage, histological type, grade), and therapeutic factors (surgical content and adjuvant chemotherapy). The robustness of these analysis is examined, using classification accuracy, 5-fold cross-validation method. Using confusion matrix, we analyze the case of misclassification.

Results: The results indicated that the deep learning model predicted with 88% accuracy on average for the test data. Misclassification occurred in stage 2 more frequently.

Conclusion: Using deep learning model, we could predict the recurrence in the early-stage endometrial cancer. More data is necessary for the accurate prediction.

キーワード

臓器別:子宮

手法別:AI

前へ戻る