Data and Scripts for 'A Predictive Model for Attention-deficit/hyperactivity Disorder and Its Subtypes in Children: A Back-propagation Neural Network Optimized with Genetic Algorithms Based on Multidimensional Cognitive Ability Tests'

Accurately identifying and diagnosing whether a child has attention-deficit/hyperactivity disorder, as well as determining the specific subtype, are crucial for selecting and implementing targeted intervention and treatment strategies. However, previous artificial intelligence-based diagnostic models for attention-deficit/hyperactivity disorder have faced challenges in achieving a balance among cost-effectiveness, low operational threshold, and high predictive accuracy. The present study recruited 316 children aged 6 to 14 years, including 98 with the inattentive subtype, 80 with the combined subtype, and 138 typically developing controls. Using multidimensional cognitive ability tests and a back-propagation neural network optimized by a genetic algorithm, this study developed a computer-aided diagnostic model to distinguish children with attention-deficit/hyperactivity disorder and its specific subtypes. The model achieved an overall predictive accuracy of 90.82%, outperforming traditional logistic regression and support vector machine models. The cancellation test was identified as the strongest predictor for differentiating attention-deficit/hyperactivity disorder subtypes. Additionally, the cancellation test, digits forward test, paper folding test, and digits backward test were critical in reflecting the cognitive ability measured by that test was more reflective of the essential differences in cognitive deficits among attention-deficit/hyperactivity disorder subtypes. Specifically, the cancellation test and digits forward test positively predicted the inattentive subtype but negatively predicted the combined subtype. In contrast, the paper folding test and digits backward test negatively predicted the inattentive subtype while positively predicting combined subtype. Moreover, the predictive contribution of cognitive tests, including the cancellation test, paper folding test, digits forward test, visual search test, and go/no-go test, varied greatly across different genders and age groups in predicting attention-deficit/hyperactivity disorder subtypes. In conclusion, the artificial intelligence-based aided diagnostic model demonstrated robust predictive performance in identifying attention-deficit/hyperactivity disorder and its subtypes in children. These findings provide a strong foundation for developing artificial intelligence-driven expert diagnostic systems for attention-deficit/hyperactivity disorder in the future.

准确识别和诊断儿童是否患有注意力缺陷多动障碍（attention-deficit/hyperactivity disorder, ADHD），并确定其具体亚型，对于选择和实施针对性干预与治疗策略至关重要。然而，以往基于人工智能（artificial intelligence, AI）的ADHD诊断模型在兼顾成本效益、低操作门槛和高预测准确性方面面临挑战。本研究招募了316名6-14岁儿童，其中包括98名注意缺陷亚型患者、80名混合型亚型患者以及138名正常发育对照儿童。本研究采用多维度认知能力测试，结合经遗传算法（genetic algorithm, GA）优化的反向传播神经网络（back-propagation neural network, BPNN），开发了一种计算机辅助诊断模型，用于区分ADHD儿童及其具体亚型。该模型的总体预测准确率达90.82%，优于传统的逻辑回归模型和支持向量机（support vector machine, SVM）模型。划消测验（cancellation test）被确定为区分ADHD亚型的最强预测因子。此外，划消测验、数字顺背测验（digits forward test）、折纸测验（paper folding test）和数字倒背测验（digits backward test）至关重要，因为这些测验所评估的认知能力更能反映ADHD亚型间认知缺陷的本质差异。具体而言，划消测验和数字顺背测验对注意缺陷亚型呈正向预测，对混合型亚型呈负向预测。相反，折纸测验和数字倒背测验对注意缺陷亚型呈负向预测，对混合型亚型呈正向预测。此外，在预测ADHD亚型时，包括划消测验、折纸测验、数字顺背测验、视觉搜索测验（visual search test）和go/no-go测验（go/no-go test）在内的认知测验的预测贡献在不同性别和年龄组间存在显著差异。综上，基于AI的辅助诊断模型在识别儿童ADHD及其亚型方面表现出稳健的预测性能。这些发现为未来开发AI驱动的ADHD专家诊断系统奠定了坚实基础。