Explainable AI for Intestinal Cancer Risk Prediction and Prevention: Case-Control Study Based on a Specific Nutritional Dataset in Marrakech-Safi Region

Abstract

Background: Intestinal cancer (IC) is among the most common cancers worldwide in both sexes. It is the third most common cancer and the second leading cause of cancer-related death globally. At the national level, IC is also highly prevalent, accounting for 49.8% of all digestive cancers, and it is the second leading cause of cancer-related death. In addition, dietary factors are critical to both the prevention and development of this cancer type; certain foods are markedly protective, whereas others confer a high risk. Objectives: The purpose of this study is to develop an intelligent system to optimize the risk of IC using a traditional Moroccan nutritional database, thereby enabling users to make personalized decisions. Methods: The database comprised 1 135 patients, of whom 493 had IC and 642 were control patients, with data on socioeconomic status, clinical history, lifestyle, behavior, and diet. It was complemented by a dietary table of traditional Moroccan dishes, together with the macronutrient composition of their respective constituents. The proposed model is based on a Transformer architecture, and SHAP (Shapley Additive exPlanations) values are used to identify risk and protective factors for personalized recommendations for the prevention of IC. Results: The Transformer model achieved an AUC (area under the receiver operating characteristic curve) of 0.984, an accuracy of 95.3%, a recall of 94.6%, a precision of 94.6%, and an F1-score of 94.7%. These values exceeded those previously reported in the literature, supporting the relevance of the proposed method. Conclusions: This study demonstrated the relevance of integrating dietary practices with contemporary predictive models of IC and the potential to develop tailored dietary guidance. This work contributes to efforts to provide effective and adaptable solutions within the Moroccan context.