Applying artificial intelligence using routine clinical data for preoperative diagnosis and prognosis evaluation of gastric cancer

Naoki Kuwayama,Isamu Hoshino,Yasukuni Mori,Hajime Yokota,Yosuke Iwatate,Takashi Uno

DOI: https://doi.org/10.3892/ol.2023.14087

2023-10-06

Oncology Letters

Abstract:The present study employed artificial intelligence (AI) machine learning technology to evaluate the prognosis of gastric cancer using blood collection data, commonly used in clinical practice and subsequently performed a stratification distinct from conventional tumor‐node‐metastasis (TNM) classification. Experiments were conducted using four machine learning methods, namely, logistic regression (LR), random forest (RF), gradient boosting (GB) and deep neural network (DNN), to classify good or poor post‐5‐year prognosis based on clinicopathological data and post‐5‐year relapse occurrence. For each machine learning method, the importance was sorted in descending order (from the most to the least); the top features were used for clustering using the k‐medoids method. The prediction accuracy and area under the curve (AUC) for 5‐year survival were as follows: LR, 76.8% and 0.702; RF, 72.5% and 0.721; GB, 75.3% and 0.73; DNN, 76.9% and 0.682, respectively. The prediction accuracy and AUC for 5‐year recurrence‐free survival were as follows: LR, 85.5% and 0.692; RF, 79.0% and 0.721; GB, 80.5% and 0.718; DNN, 83.2% and 0.670. Clustering patients into three groups resulted in a stratification distinct from the TNM classification. In conclusion, AI machine learning using routine clinical data can help evaluate the prognosis of gastric cancer, with prognosis differing according to AI‐identified clusters.

oncology