演題抄録

ポスター

開催概要
開催回
第56回・2018年・横浜
 

人工知能圧縮センシングを用いた生存率解析理論における誤差の検討

演題番号 : P30-1

[筆頭演者]
演者)宮木 康成:1 
[共同演者]
小田 隆司:2、三宅 貴仁:2、藤原 恵一:3

1:三宅おおふくクリニック・婦人科、2:三宅医院・産婦人科、3:埼玉医科大学国際医療センタ-・婦人科腫瘍科

 

The relationship between the dimensions of the original vector and the adding vector for estimating the values of the clinical sample by the compressive sensing with L1 regularization of the artificial intelligence for survival analysis was investigated.
When m>n (m and n are natural number), X= (x1,...,xm); (xi is not negative number, i=[1,m]), Y= (y1,...,yn); (yj is not negative number, j=[1,n]), A(ap,q); ap,q is consisted of the elements of standard normal distribution, (p=[1,n], q=[1,m]), Y is known, and argmin (||Y-AX||2)/n+λΣ|xi| under all elements of Y are included in X, X is likely acquired as the predicted vector. Then, when n=[3,15], m=[n+2, n+2.5n], Y=[n,m] and d is the error vector, z1=sqrt(Σd2)/n was repeatedly obtained.
When the values were transformed as s=n/(n+m) and z=(ln(z1)-min(ln(z1)))/(max(ln(z1))-min(ln(z1))), the transformed error value z showed a good fit as a function of s, following z=1/(1+e-14.85s+7.66) (p<10-19).
The bigger m became, the smaller z became. When this procedure is applied to clinical studies, the more precise prediction will be obtained if the dimension of the adding vector is bigger than twice of the original vector.

キーワード

臓器別:卵巣

手法別:Clinical Trial (臨床試験)

前へ戻る