Clinical Gastrointestinal Disease Screening Based on the Bionic Artificial Tentacle Method

Background

Gastrointestinal diseases display a wide range of complex symptoms globally, such as diarrhea, gastrointestinal bleeding, malabsorption, malnutrition, and even neurological dysfunction. These diseases pose significant health challenges and socioeconomic burdens due to their noticeable regional, age, and gender differences, especially gastrointestinal cancers, which account for one-third of global cancer incidence and mortality. Early screening and timely intervention of gastrointestinal diseases are essential to reduce mortality and improve life expectancy.

Traditional gastrointestinal screening primarily relies on endoscopy, using flexible endoscopes with embedded cameras to inspect the gastrointestinal tract through natural openings. Despite its widespread clinical use, endoscopy still has limitations with optical sensors, poor illumination conditions, and highly constricted working environments in the gastrointestinal tract. These factors can lead to darkness, glare, reflections, reducing image quality, and potentially causing misdiagnosis. Wireless capsule endoscopy, which captures images non-invasively, has garnered significant interest. Nonetheless, traditional endoscopy is still required for detailed diagnosis following capsule endoscopy.

In current clinical practice, tactile sensing modes have not yet been applied, limiting doctors’ ability to identify small polyps and any abnormality in tissue structure and hardness caused by diseases. Recent studies have shown that rats gather tactile information about their environment via their whiskers, which can complement visual methods by providing tactile feedback. Based on the whisker sensing mechanism of rats, we propose a hardware system based on artificial whiskers aimed at gastrointestinal screening applications. This system combines artificial intelligence and self-learning capabilities to recognize tissue structures and pathological changes through tactile pattern recognition, providing a new method for clinical screening and reducing diagnostic discrepancies among physicians.

Paper Source

This paper is co-authored by Zeyu Wang, Frank P.-W. Lo (corresponding author), Yunran Huang, Junhong Chen, James Calo, Wei Chen, and Benny Lo (corresponding author), from the Department of Surgery and Cancer at Imperial College London, Fudan University Shanghai Medical College, and the Center for Intelligent Medical Electronics at Fudan University. The paper was published in 2023 in the journal npj Robotics.

Research Process

The study followed these main steps:

1. Concept Design

The research team designed a hardware system based on bionic artificial tentacles for gastrointestinal inspection. The system’s working mechanism and principles are shown in Figure 1: By simulating the sensing mechanism of rat whiskers, the system can identify pathological abnormalities on the gastrointestinal wall, obtain structural and contour information of diseased tissues, and aid clinical decision-making.

2. Modeling and Design Parameter Optimization

The research team modeled the sensing mechanism of the tentacles, comparing the working principles of static and dynamic models. Figure 2 shows these two models and analyzes the advantages and disadvantages of each. For instance, while the static model is suitable for positioning accuracy tasks, it has slow detection speed and significant challenges in system noise control. In contrast, the dynamic model enhances sensing capability through external drive, improving integration and system robustness.

3. System Design and Implementation

Based on the model comparisons above, the research team designed a low-noise, fast-response, and wide dynamic range tentacle hardware system, including tentacle sensors and signal conditioning circuitry. The tentacle sensor, constructed from polyvinylidene fluoride (PVDF) film sensors, offers excellent signal conduction performance and flexibility. The signal conditioning circuitry, through multi-stage amplification, filtering, and analog-to-digital conversion processes, translates raw electrical signals into high-fidelity digital signals, further transmitted to a computer for analysis and interpretation.

4. Benchmark Testing and Core Function Evaluation

To assess the electrical performance of the system, the research team conducted a series of benchmark tests, including inherent noise testing, texture recognition, distance perception, hardness characterization, and shape recognition experiments. The experimental results showcased the system’s performance in different tasks, such as time-domain and frequency-domain response characteristics of output signals. Specific tests are detailed in Figures 4 to 7.

Research Results

Inherent Noise Evaluation

Noise testing results within a temperature range showed stable inherent noise, with a maximum of 0.75 uVrms and 6.16 uVpp, and minimal thermal effects on signal quality.

Texture Recognition

Significant differences were observed in signal responses in the time domain, frequency domain, and time-frequency domain for different surface materials. For instance, smooth surfaces corresponded to smooth signal outputs, while rough surfaces like sandpaper produced more high-frequency components.

Distance Perception

Experimental results for different height settings showed a highly linear correlation between the tentacle system’s signal release point and the height parameter, with a good fit.

Hardness Characterization

Significant signal response differences were observed between soft tissue and bone tissue, indicating the system’s effectiveness in distinguishing material hardness.

Shape Recognition

The tentacle system demonstrated significant capability in shape recognition through experiments on circular, plane, and inclined surfaces.

Clinical Application Feasibility

To evaluate the clinical application potential of this method, the research team conducted a preliminary study using simulation models (see Figure 8) to detect three typical diseased tissues: normal tissue, ulcerative colitis, and ulcerative cancerous tissue. Results from 120 experiments indicated that the model trained by the deep learning algorithm could efficiently and accurately identify diseased tissues, achieving a test accuracy of 94.44% and a kappa value of 0.9167.

Conclusion and Significance

This study proposes a novel gastrointestinal screening method based on artificial tentacles. By combining hardware design and deep learning algorithms, it shows great potential in extracting structural and textural information from the gastrointestinal tract, potentially complementing or enhancing existing visual endoscopy diagnostic techniques. The system’s highly integrated hardware design and low computational power characteristics suggest this method could serve as an independent submodule of existing endoscopic robotic surgery platforms, enabling automated sensing, analysis, and diagnosis with minimal or no manual intervention, reducing discrepancies among doctors, and serving as a low-cost early screening mechanism in developing countries.

Further research in multi-channel hardware system design and algorithm optimization may allow this new method to be fully developed and clinically applied. The research team also plans to explore new sensing solutions, such as strain gauges, to further enhance the static sensing capabilities of the tentacle system.