A Study of Predicting the New Prognostic Evaluation of Gastrectomized Patients with Cancer of the Stomach Using a Neural Network Model

Takamasa Hiraoka, Takao Umemoto, Katuyuki Kunieda,　Motohisa Kato, Iwao Kumazawa and Shigetoyo Saji

Second Department of Surgery, Gifu University School of Medicine

The prognosis of 261 gastrectomized pasients (Stage I: 117, II: 31, III: 62, IV: 51) out of 411 gastric cancer pasients for 5 years starting from 1985 were statistically analyzed according to the prognostic factors of Japanese classification of gastric carcinoma and serum levels of pre-and postoperative tumor markers, by uni-and multivariated methods. From the above results, the personal prognosis was predicated using a neural network. This network was constructed with 9 neurons for the input layer information such as Stage, serum CEA and IAP levels at one month after operation, the difference of pre-and post-operative CEA value, curative and non-curative resection, number of metastatic lymph nodes, 9 neurons for the middle layer and 2 neurons for the output layer. After learning this neural network by the back propagation method, 94 out of 130 pasients (72.3%) satisfactorily gave the correct answer. Also, using the learned neural network, the prognosis of 90 out of 131 unknown patients (68.7%) was predicted correctly. In terms of the treatment effects, the neural network model may be useful for predicting the new personal prognosis of gastrectomized pasients.

Key words
neural net work, gastric cancer, postoperative prognosis, clinical statistical analysis

Reprint requests
Takamasa Hiraoka Second Department of Surgery, Gifu University School of Medicine 40 Tsukasa, Gifu, 500-8705 JAPAN

Accepted
March 31, 1999

To read the PDF file you will need Abobe Reader installed on your computer.