Objective To construct a multimodal model integrating blood routine scatter plots and clinical and laboratory examination indicators for the early clinical auxiliary diagnosis of influenza A.

Methods The clinical manifestations, blood routine/biochemical indicators and blood routine scatter plots of 580 children with respiratory tract infections were collected. The stratified sampling was used to identify divide the training set, validation set and test set. Table, image and multimodal fusion models were established, respectively. The performance of the models was evaluated by the accuracy, sensitivity, specificity, F1 score and AUC. SHAP was used to explain the key features of the models.

Results The accuracy rate of the multimodal model in the independent test set was 88.8%, and the AUC was 0.922, which was superior to the table-only model (accuracy rate 77.6%, AUC = 0.831) and image-only model (accuracy rate 80.2%, AUC = 0.905). The results of SHAP analysis showed that the lymphocyte count, white blood cell count, procalcitonin, lymphocyte/monocyte ratio and neutrophil count were the main contributing characteristics.

Conclusions This study constructs an auxiliary diagnostic model for influenza A in children that integrates blood routine scatter plots and clinical laboratory indicators, which can enhance diagnostic efficiency, and provide a reference for rapid screening at the grassroots level.