Electronic Theses and Dissertation

Karies gigi merupakan masalah kesehatan global dengan lebih dari dua miliar kasus aktif, namun deteksi dininya masih terhambat oleh keterbatasan akses layanan kesehatan, khususnya di wilayah dengan infrastruktur jaringan yang buruk. Aplikasi Gigoe Detection V.01 memiliki tiga kelemahan utama: hanya mendukung platform Android, bergantung penuh pada koneksi internet untuk proses inferensi, dan mendapat evaluasi negatif dari pengguna terkait antarmuka. Penelitian ini bertujuan redesign aplikasi tersebut dengan mengimplementasikan kapabilitas cross-platform (Android dan iOS), arsitektur offline-first, serta peningkatan antarmuka berbasis hasil usability testing sebelumnya. Penelitian menggunakan pendekatan pengembangan perangkat lunak dengan framework Flutter dan bahasa Dart, mengintegrasikan model YOLOv8 TFLite secara lokal (on-device AI), Firebase sebagai backend, serta pipeline CI/CD GitHub Actions untuk menghasilkan build iOS. Pengujian dilakukan melalui Black Box Testing, System Usability Scale (SUS), pengukuran latensi end-to-end, dan analisis konsumsi daya menggunakan metode delta battery percentage terhadap 11 responden dokter gigi. Hasil penelitian menunjukkan seluruh modul fungsional berjalan 100% sesuai rancangan, skor SUS meningkat menjadi 79,09 (Grade B), latensi on-device inference mencapai 4,78 ms lebih rendah dibandingkan cloud inference dan arsitektur offline-first terbukti menjaga integritas data tanpa koneksi internet. Penelitian ini membuktikan bahwa pendekatan cross-platform berbasis Flutter dengan arsitektur offline-first mampu menghadirkan aplikasi mobile health deteksi karies yang andal, efisien, dan dapat digunakan di berbagai kondisi operasional lapangan.

Dental caries is a global health problem with more than two billion active cases, yet early detection remains hindered by limited healthcare access, particularly in areas with poor network infrastructure. The Gigoe Detection V.01 application had three major limitations: it only supported Android, relied entirely on internet connectivity for inference, and received negative user feedback regarding its interface. This study aims to redesign the application by implementing cross-platform capability (Android and iOS), an offline-first architecture, and interface improvements based on prior usability testing results. The study employs a software development approach using the Flutter framework and Dart language, integrating a local YOLOv8 TFLite model (on-device AI), Firebase as the backend, and a GitHub Actions CI/CD pipeline to generate iOS builds. Testing was conducted through Black Box Testing, System Usability Scale (SUS) evaluation, end-to-end latency measurement, and battery consumption analysis using the delta battery percentage method with 11 dentist respondents. Results show that all functional modules performed 100% as designed, the SUS score improved to 79.09 (Grade B), on-device inference latency reached 4.78 ms lower than cloud inference and the offline-first architecture successfully maintained data integrity without internet connectivity. This study demonstrates that a Flutter-based cross-platform approach with offline-first architecture can deliver a reliable and efficient mobile health application for dental caries detection across diverse operational conditions.