**Background:** Obesity prevalence has tripled since 1975, with strong links to hypertension, cardiovascular disease, diabetes, and mood disorders. Psychological and behavioral interventions are effective but often require intensive in-person sessions, creating access barriers. Technology-based interventions have shown promise for weight loss through improved adherence and self-monitoring, but few incorporate mood and eating behavior components. The CogniNU app was developed to integrate artificial intelligence (AI), cognitive behavioral therapy (CBT), and nutritional science into a single smartphone platform.
**Methods:** This single-center prospective, randomized, crossover pilot study (ISRCTN16082909) was conducted at Kaohsiung Medical University Hospital. Twenty adults aged 30–60 years with BMI ≥ 25 kg/m² were enrolled (median age 42.0 years, IQR 36–47.5; 75% female; median BMI 27.9 kg/m², IQR 25.9–33.3). Exclusion criteria included major neuropsychiatric comorbidities. Participants were randomized 1:1 to control-first or intervention-first groups. The intervention group used the CogniNU app for 30 days, which employed AI food image recognition (based on >370,000 food images from nine open-source datasets), ingredient categorization, and nutrient analysis linked to a CBT-based digital program. After 30 days, groups crossed over. Body composition was measured using InBody 770 bioelectrical impedance analysis. Behavioral and psychosocial outcomes were assessed via the Dietary Behavior Questionnaire (DBQ), Mindful Eating Behavior Scale (MEBS), Brief Symptom Rating Scale (BSRS-5), Beck Depression Inventory II (BDI-II), WHOQOL-BREF, International Physical Activity Questionnaire Short Form (IPAQ-SF), and Pittsburgh Sleep Quality Index (PSQI). Nonparametric tests were used (Wilcoxon rank sum, Wilcoxon matched-paired sign-rank, Fisher's exact test).
**Key Results:** In the control-first group, body weight changed by +0.8 kg during the control period versus −0.6 kg during the intervention period (p=0.02), and body weight change percentage was +0.7% versus −0.6% (p=0.02). In the intervention-first group, body fat percentage decreased by 0.4% during the intervention period and increased by 0.05% during the control period (p=0.01). When data were integrated into App (n=20) and As usual (n=20) groups, the App group showed a marginal improvement in MEBS total scores (Δmean=0.94 vs. −1.7, p=0.07) and a statistically significant improvement in the 'eating in response to hunger and satiety cues' domain (Δmean=0.7 vs. −0.61, p=0.04). No significant changes were observed in mood (BSRS-5, BDI-II), sleep quality (PSQI), physical activity (IPAQ-SF), or quality of life (WHOQOL-BREF) between groups. No app-related adverse events were reported. The retention rate was 100%.
**Clinical Implications:** This pilot study provides preliminary evidence that a 30-day AI-integrated CBT-based smartphone app can produce modest but statistically significant improvements in body weight (−0.6 kg), body fat percentage (−0.4%), and mindful eating behavior among overweight/obese adults. The improvement in hunger and satiety cue responsiveness suggests the app may help users develop healthier eating patterns. The high retention rate (100%) and absence of adverse events support feasibility and safety. However, the small sample size (n=20), short intervention duration, homogeneous population (hospital employees), and lack of significant effects on mood or quality of life limit generalizability. The authors estimate that at least 49 participants per group would be needed for a definitive trial. These findings support further investigation of technology-based comprehensive weight management interventions that address both behavioral and psychological components.