This software implements the methods from the CVPR 2026 research paper titled Similarity-as-Evidence. Computer models often guess wrongly when they feel overconfident in medical exams. This application calibrates these models to prevent errors. It provides evidence for why a model makes a specific choice. This approach helps doctors and researchers trust automated feedback. You use it to label medical images with less work and higher precision.
Your computer needs specific components to run this application smoothly. Ensure you meet these standards before you begin:
- Operating System: Windows 10 or Windows 11.
- Processor: A modern multi-core processor (Intel Core i5 or AMD Ryzen 5 or better).
- Memory: At least 8 gigabytes of RAM.
- Storage: 5 gigabytes of free disk space for program files and cached medical image sets.
- Graphics: A dedicated graphics card is recommended for faster image processing.
To begin, visit the project page to download the software.
Click here to visit the project repository page
Follow these steps to set up your environment:
- Open the link provided above in your web browser.
- Locate the green button labeled Code.
- Choose Download ZIP from the menu.
- Save the file to a folder on your computer.
- Extract the contents of the ZIP folder to a space on your hard drive.
- Open the folder once the extraction completes.
The software requires a few setup steps to prepare your computer. Please follow this guide:
- Find the file labeled setup.bat in the main folder.
- Double-click the file to start the automated process.
- A window will appear and show the progress. Keep this window open until it tells you the setup is complete.
- The system downloads necessary data libraries automatically. This may take several minutes depending on your internet speed.
- Once the setup finishes, press any key to close the window.
You can start the program now that the setup is complete.
- Find the file named run_application.bat in the same folder.
- Double-click this file.
- Your default web browser will open to the software interface.
- Navigate to the Upload tab to select your medical image files.
- Select a folder of images or individual files.
- Click the Analyze button to begin the similarity process.
- The software displays the evidence for each image prediction in the results view.
The application groups images based on their similarity to known medical samples. When the model looks at your images, it performs a comparison. If the model finds a strong match, it displays the corresponding sample. You see the similarity score next to each result. A higher score means the system is more certain about its choice. If the score is low, the system flags the image for your manual review. This process saves you hours of time by focusing your attention only on difficult cases.
Most users experience few problems during operation. If the application stops responding, try these steps:
- Restart the program: Close the window and run the run_application.bat file again.
- Check internet status: The program performs occasional checks for updates. Ensure you have an active connection.
- Confirm file access: Ensure the files are not located in a protected system folder. Store them in your Documents or Downloads folder for the best results.
- Verify file types: The software currently supports PNG and JPG formats for medical images. Convert files to these formats if the software does not recognize them.
- Free space: If analysis stops halfway, check if your computer has enough free disk space to store the results.
The software saves all results in a subfolder named Output. Each time you run an analysis, the program creates a new text file with the date and time. This keeps your work organized. You can move or rename these files as you see fit. The software never deletes your original medical images. It only reads them to create the similarity reports.
- Use high-resolution images for the best similarity scores. Poor quality images can lead to incorrect matches.
- Keep your image filenames consistent to make tracking results easier.
- If you process large batches of images, perform the work in smaller groups of 50 to 100 files to maintain system speed.
- Regularly clear the temporary files in the Cache folder if you notice the application slowing down over time.
This software runs entirely on your local machine. No images or data are sent to external servers or the cloud. You keep complete control over your medical data. This keeps the information secure and compliant with local data handling habits. You can use the software while offline to ensure your files remain contained within your local hardware. Feel free to disconnect from the internet after the initial setup process is complete.