Development of an AI-Based Smartphone Application for Rapid Tick Identification and Geospatial Mapping: A Pilot Study

The study built a two-stage AI pipeline that first localizes tick regions with a YOLO-based model and then performs genus-level classification with deep learning models, integrated into a prototype smartphone application with geospatial visualization. The text describes it as the first AI-based smartphone application in Japan for automated tick identification and geospatial visualization, linking detection, classification, and map display in one prototype. The detection model was trained on 3258 annotated public images and achieved mAP@0.5 of 0.954; classification used 533 tick images covering four genera, with ResNet50 reaching mean validation accuracy of 95%.

The classification model achieved high validation accuracy despite substantial class imbalance, and visualization analyses indicated the model primarily focused on tick morphological features rather than background elements. The text notes that wild tick images were added to address class imbalance and that visualization analyses examined where the model focused, rather than reporting overall accuracy alone. The classification dataset comprised 533 tick images across four genera; the text reports mean validation accuracy of 95% for ResNet50 and states that visualization analyses confirmed focus on tick morphological features.

The application is positioned as potentially supporting clinical risk assessment, patient counseling, reduction of unnecessary empirical antibiotic use, and enhanced public health surveillance. The text connects genus-level identification to clinical relevance by noting that pathogen transmission varies by tick genus while tick identification is rarely performed in routine clinical practice. These are proposed potential applications phrased as 'may support'; the text does not report measured clinical outcomes or changes in antibiotic use.

The text limits the scope to genus-level identification and microscope-acquired imaging, with the system presented as a prototype smartphone application with geospatial visualization; its intended use is to support clinical risk assessment, patient counseling, reduction of unnecessary empirical antibiotic use, and public health surveillance. For readers, it offers a technical path that combines tick image identification with geospatial display, relevant to researchers, clinicians, and surveillance personnel interested in tick-borne disease applications.

As a pilot study, the text does not report clinical outcomes, changes in antibiotic use, or public health surveillance effects; classification is limited to genus level and images are microscope-acquired, so performance under real smartphone capture conditions remains to be observed; wild tick images were added to address class imbalance, and generalization across different regions and tick species compositions remains an open question.