History of Updates
Changelog of application changes and model experiments
Latest App Update: October 24, 2026
Latest Model Update: January 27, 2026
Application Updates
Live Counter Board on the Dashboard
major
v2.9.0•October 24, 2026Added a live counter board to the dashboard that updates automatically as scans are processed
Changes:
- •Added a counter board showing scans processed, brain tumors found, AI models online and typical analysis time
- •Counters refresh automatically every 15 seconds, so newly processed scans appear without reloading the page
- •Every tumor the models flag in a scan is counted, so one scan can add more than one tumor
- •AI Models Online shows the result of the latest automatic model check
- •Typical Analysis Time shows the median time the models took on the most recent scans
Research & Learning Navigation
minor
v2.8.0•October 15, 2026Made the project's research and learning pages easier to find from anywhere on the site
Changes:
- •Added links to About the Project, MRI & DWI learning, Model Transparency and History of Updates in the front page header, with a menu on small screens
- •Added a Research & Learning section to the front page
- •Added Research & Learning links to the dashboard and scanner pages
- •Added related research links at the end of the About, MRI Education, Model Transparency and History of Updates pages
- •Updated third-party software components to fix known security issues
Security & Privacy Hardening
security
v2.7.1•September 25, 2026A further round of security and privacy work across the site
Changes:
- •Strengthened protection of stored scans and reports
- •Strengthened administrator sign-in protections
- •Added authorization checks to more server features
- •Reduced the information kept in server logs to protect privacy
- •Updated third-party software components to fix known security issues
Accessibility & Public Page Improvements
minor
v2.7.0•September 25, 2026Made the site easier to use with a keyboard or screen reader, and made public pages load faster
Changes:
- •The MRI upload area and the main navigation can now be used fully with a keyboard
- •Improved the heading structure of public pages so screen readers can move through them
- •The MRI Education, Model Transparency and Help Center pages are now pre-built, so their content appears right away, even without JavaScript
- •Each public page now loads with its own title and description
- •Missing files now return a proper "not found" response
Reframed as an Experimental Student Research Project
major
v2.6.0•September 25, 2026The site now presents itself clearly as an experimental student research project for education and research, not a diagnostic tool
Changes:
- •Rewrote wording across the site, search listings and PDF reports to describe the project as experimental student research
- •Model metrics are now labeled as development metrics from each model's validation split, with a note that results on real uploaded scans are substantially lower
- •Added a "Development metrics vs. real-world scans" section to the Model Transparency page
- •Added a research journey section to the About page and a Help Center question on why results differ from the published model metrics
- •Results screens and PDF reports now describe findings as model output instead of diagnostic statements
- •Improved the PDF report layout so text no longer overlaps
Security & Privacy Improvements
security
v2.5.1•June 10, 2026Fixes from a security review of the site
Changes:
- •Strengthened access controls for scan results, images and reports
- •Strengthened protection of administrator features
- •Improved how the server handles files and service settings
- •Scan images in the review tools now load through a secure viewer
- •Updated third-party software components to fix known security issues
Search Visibility & Page Structure
major
v2.5.0•June 10, 2026Helped search engines, link previews and AI assistants understand the site's public pages
Changes:
- •Each public page now gives search engines and AI assistants its own title, description and summary
- •Marked up the Help Center questions and answers so search engines can understand them
- •Clearer page structure with one main heading per page; the Terms of Service and Privacy Policy are now split into sections
- •Added a preview image for links shared on social media
- •Unknown addresses now return a proper "not found" response, and /home now redirects to the dashboard
- •Corrected the Help Center to state the actual 5 MB upload limit
Broader Incorrect-Result Reporting
minor
v2.4.0•February 22, 2026More results can now be reported as incorrect, which helps show where the models go wrong
Changes:
- •Results where the model flagged nothing can now be reported, for example when a tumor was missed
- •Flagged results can now be reported when model confidence is below 95% (previously below 85%)
DICOM File Uploads
major
v2.3.0•February 6, 2026The scanner now accepts DICOM files, the standard file format for medical images such as MRI scans
Changes:
- •DICOM (.dcm) MRI files can now be uploaded and analyzed, alongside PNG and JPEG images
- •DICOM files are converted to a standard image automatically before analysis
- •Supports uncompressed and common compressed DICOM files (JPEG, JPEG 2000 and RLE)
Landing Page Redesign for the Three AI Models
major
v2.2.0•January 28, 2026Redesigned the front page to introduce all three AI models
Changes:
- •Redesigned the landing page to present Original Detection, Tumor Classification (Object Detection) and Tumor Classification (Segmentation)
- •Corrected the model metrics on the landing page to match each model's validation results
- •Improved the readability of buttons, step numbers and headings on the landing page
- •Updated the System Status page with a status indicator for each AI model and a live countdown to the next automatic check
- •Updated links and the sitemap to use the site's main address, ai-tumor-detection.org
Automatic AI Model Health Monitoring & Multi-Model Support
major
v2.1.0•January 27, 2026Infrastructure update introducing automatic model testing, multi-model selection, and security changes
Changes:
- •Implemented automatic AI model health monitoring with 3-hour testing intervals
- •Added server-side scheduler that tests all 3 AI models without user intervention
- •Created read-only System Status page displaying model health, response times, and next check countdown
- •Introduced model selection feature allowing users to choose between 3 different AI models before upload
- •Removed hardcoded API key fallbacks - enforces environment variable configuration
- •Updated system dependencies (jspdf@3.0.3, preact@10.27.0, qs@6.13.0) for improved stability
- •Added model status API endpoint with cached results to minimize API costs
Legal Documents & User Experience
major
v1.0.0•August 10, 2025Major update focusing on legal compliance and improved user interface
Changes:
- •Added Privacy Policy with automatic GMT date formatting
- •Implemented detailed Terms of Service with user responsibilities and liability protections
- •Created real-time Status Page with health monitoring for all system components
- •Updated Help Center with detailed FAQ and tutorial content
- •Improved medical disclaimer system with mandatory acceptance flow
- •Updated contact integration across all pages with functional email links
AI Model Updates
Triple AI Model System - YOLO26 Classification Models
major
v3.0.0•January 27, 2026Added two classification models alongside Original Detection for experimenting with different model outputs
Improvements:
- •Included Original Detection (RF-DETR v2), which draws boxes around regions it flags as possible tumor
- •Introduced Tumor Classification Object Detection (YOLO26), which draws boxes and predicts a tumor type
- •Implemented Tumor Classification Segmentation (YOLO26), which draws pixel outlines and predicts a tumor type
- •YOLO26 models trained on 9,711 MRI scans augmented to 40,606 training samples
- •Classification models predict one of three tumor types: Glioma, Pituitary, Meningioma
- •Created dedicated results page for classification models showing tumor type information
- •Implemented automatic model health testing every 3 hours to minimize API costs
RF-DETR v2 Model Upgrade - RF-DETR (Medium)
major
v2.0.0•August 20, 2025Moved Original Detection to the RF-DETR (Medium) architecture
Development metrics are measured on the model's held-out validation split of its own development dataset. Real-world performance on uploaded scans is substantially lower. Why real-world results are lower.
Improvements:
- •Migrated to RF-DETR (Medium) architecture
- •Upgraded to new 'version2' inference workflow
- •Development metrics on the held-out validation split: mAP@50: 97.9%, Precision: 97.6%, Recall: 97.0%
- •Trained the model on an expanded dataset for tumor region detection
- •Optimized API endpoints for faster inference and better reliability
- •Maintained backward compatibility with existing 'object' to 'Tumor' label conversion
Detection Workflow Integration
major
v1.0.0•August 10, 2025Changes to the detection workflow and processing speed
Improvements:
- •Upgraded to 'detect-count-and-visualize' workflow for region detection
- •Updated tumor boundary detection
- •Updated confidence scoring algorithm
- •Optimized processing speed reducing analysis time by 40%
- •Added support for detecting multiple tumor types and sizes
- •Implemented image annotation with region labels