Predicting the Prognosis of the Hepatectomized Patient with Hepatocellular Carcinoma with a Neural Network

Isao Hamamoto, Setsuo Okada, Hisao wakabayashi, Takashi Maeba, Hajime Maeta

First Department of Surgery, Kagawa Medical School

The prognosis of the hepatectomized patients with hepatocellular carcinoma was predicted with a neural network. A neural network with 9 neutons for the input layer, 14 neurons for the middle layer and1 neuron for the output layer was constructed. Preoperative data (AST,ALT, ALP, hepaplastine test, ICGR 1.5, total liver volume, residuall liver volume, platelet and total bilirubin) of 58 hepatectomized patients (49 and 9 patients were discharge from and died at hospital, respectively) whose prognoses were already known, were learned by the neural network with teaching signals (1 for discharged, 0 for hospital death). After 100,000 times of learning, the output of the neural network converged satisfactorily and gave the correct answer in all except one case. With the learned neural network, the prognoses of 11 patients with hepatocellular carcinoma were estimated prospectively. The prognoses of these patients, 10 of them were discharged and one of whom died of hepatic dysfunction, were predicted correctly. It was concluded that the neural network was a powerful tool for predicting the prognosis of hepatectomized patients with hepatocellular carcinoma.

Key words
neural network, hepatocellular carcinoma, hepatectomy

Reprint requests
Isao Hamamoto First Department of Surgery, Kagawa Medical School
1750-1 Ikenobe, Miki-cho, Kita-gun, Kagawa, 761-07 JAPAN

Accepted
February 8, 1995

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