Restore research disclaimer to UI

Remove research disclaimer from UI

Add viewport meta tag and fix mobile UI to fill entire screen with fixed positioning

Fix mobile UI to cover full screen height on mobile

Disable CLIP validation - model works without it, PACS pre-filters images

Simplify CLIP validation - only require hand_xray_score > non_hand_score

Lower CLIP validation threshold from 0.40 to 0.30

Add debug mode to CLIP validation - log individual scores for each description

Make mobile UI cover full screen width and height

Add comprehensive timing instrumentation to measure inference bottlenecks

Significantly improve mobile UI sizing

Add DICOM image support

Make Assess Bone Age button more visible

Improve mobile responsiveness

Simplify validation error message

Change error message text to golden color

Switch error message to white text on dark red background

Make error text much darker and bolder for readability

Improve error message readability and actionability

Add file size limits and improve error message visibility

Add CLIP-based image validation to reject non-hand X-rays

Update page title to 'Bone age AI tool'

Update page title to 'Kao-Sato Bone age assessment AI tool'

Reduce modal font sizes for compact view

Improve modal readability: better contrast and larger text

Add confidence indicator info modal with legend

Fix color coding: use !important to override dark mode defaults

Add color-coded confidence bar: green (high) / orange (medium) / red (low)

Fix: use_deterministic_algorithms() instead of set_deterministic()

Re-enable TTA for accuracy over speed

Ensure deterministic inference

Simplify inference: remove padding & TTA, match training exactly

Remove auto-invert heuristic for normal PACS data

Fix inference pipeline: fp32 on CPU + thread-safe CLAHE

Add CLAHE adaptive histogram equalization for contrast normalization

Add EXIF orientation fix and auto-crop to hand ROI

Submit form on Enter key when image is loaded

Restore model uncertainty label (±8 months)

Remove 2SD display until values are verified

Show -2SD and +2SD on separate lines in result

Show GP ±2 SD normal range in result

Improve inference: TTA, aspect-ratio padding, auto-inversion, prediction clipping

Add collapsible image tips panel to upload field

Add percentile contrast normalization for screenshot robustness

Dark mode: teal-forest palette with radial gradient bg

Fix dropzone text invisible in dark mode

Remove KaoNet name from UI header

UI aesthetic overhaul — KaoNet design system

Rebrand UI to KaoNet

Handle model load errors gracefully