An open-source, real-time colonic polyp detector that runs on an ordinary Android smartphone, with no special hardware and no internet connection.
Research use only. This app is a research prototype. It is not a medical device, has not been approved or cleared by any regulatory authority, and must not be used for clinical diagnosis or patient care decisions.
This repository accompanies:
Kim YB. Smartphone-based polyp detection: a first step towards an open-source AI framework. J Med Artif Intell. 2025;8:49. doi:10.21037/jmai-24-310
The work was first presented as an invited on-site poster at ESGE Days 2023 ("Home-Made Polyp-Detector: Deep-Learning Based Polyp Detector Using Android Smartphones").
If you use this software, please cite the paper above. GitHub's "Cite this repository" button uses CITATION.cff.
This application is a modification of the TensorFlow Lite object detection example app, a camera app that continuously detects and identifies general objects. This modified app detects and identifies colonic polyps using the device's back camera.
The polyp models are quantized EfficientDet-Lite2 detectors trained on the following public dataset (CC0 1.0):
- Wang, Guanghui, 2021, "Replication Data for: Colonoscopy Polyp Detection and Classification: Dataset Creation and Comparative Evaluations", Harvard Dataverse, V1. doi:10.7910/DVN/FCBUOR
For optimal performance, run the app on a physical Android device.
-
The Android Studio IDE. This sample was tested on Android Studio Bumblebee.
-
An Android device with a minimum OS version of SDK 24 (Android 7.0 - Nougat) with developer mode enabled. The process for enabling developer mode can vary by device.
-
Clone this repository.
-
Launch Android Studio. From the Welcome screen, select "Open an existing Android Studio project."
-
In the "Open File or Project" window, select the root folder of this repository. Click OK.
-
If prompted for a Gradle Sync, confirm by clicking OK. Android Studio creates
local.propertieswith your SDK path automatically; this file is not committed. -
Connect your Android device to your computer with developer mode enabled, then click the green Run arrow to build and deploy the app.
The polyp detection models are bundled in app/src/main/assets. Select one from the "ML Model" menu in the app:
| File | Description |
|---|---|
efficientdet-lite-polyp.tflite |
Polyp detector (EfficientDet-Lite) |
efficientdet-lite2-polyp.tflite |
Polyp detector (EfficientDet-Lite2) |
efficientdet-lite2-selected.tflite |
Polyp detector (EfficientDet-Lite2, variant) |
efficientdet-lite2-camera.tflite |
Polyp detector (EfficientDet-Lite2, variant) |
efficientdet-lite2-high-camera.tflite |
Polyp detector (EfficientDet-Lite2, variant) |
The general-purpose TensorFlow models (MobileNet v1, EfficientDet-Lite0/1/2) are downloaded automatically at build time by app/download_models.gradle.
The code is licensed under the Apache License 2.0, the same license as the TensorFlow example it is based on. See NOTICE for attributions and third-party components. The training dataset is in the public domain under CC0 1.0 and is not redistributed here.