AI-Enhanced Wearable Antenna System for Wireless Capsule Endoscope Localization in Heterogeneous Environments

Abstract

Wireless capsule endoscope (WCE) localization is critical for targeted drug delivery and precise diagnosis of gastrointestinal (GI) lesions. However, the precise location and reliable tracking of a WCE within the complex anatomical environment of the human body remain a significant challenge. This study investigates an AI-integrated wearable antenna system for WCE localization in heterogeneous environments. The experimental validation is conducted using a realistic multilayer heterogeneous phantom (MHP) that replicates human tissue properties, comprising distinct skin, fat, and muscle layers, with an inner cavity filled with GI tract-mimicking saline solution. The proposed system employs a conformal capsule transmitter (Tx) and four wearable receiver (Rx) antennas operating at 2.45 GHz in the industrial, scientific, and medical band, which are interfaced with software-defined radios (SDRs) to collect received signal strength (RSS) data. The measurement setup enabled the collection of 3300 RSS data points, which were systematically mapped to determine the capsule’s position. Data-driven frameworks employing five machine-learning (ML) regression models: k-nearest neighbors (KNNs), adaptive boosting (AdaBoost), decision tree (DT), radio frequency (RF), and extreme gradient boosting (XGBoost), were utilized to enhance localization accuracy. Among these, XGBoost demonstrated the highest localization accuracy, with a remarkable root-mean-square error (RMSE) of 1.69 cm for 3-D localization and 0.39 cm for trajectory tracking within the MHP. These findings highlighted the potential of the proposed AI-integrated wearable antenna system for localization and precise real-time capsule tracking, thereby advancing WCE-based diagnostics and treatment strategies.