Remove skull stripping and slice filtering from preprocessing pipeline

Update loading copy: replace BrainFOCUS AI and AI model references with System

Clean layout: move upload title and support format inside the UploadZone box

Update footer text to sound more professional and standard for medical referencing

Convert Demo Case 2 to run real AI model on mri-02_stripped.nii (Normal case)

Clean up upload zone: remove static helper text and keep only dynamic drag/drop statuses

Simplify demo cards: remove icons and description text, keep only centered headers

Update homepage copy: revise titles, upload labels, and demo card headers

Force equal height columns: apply CSS Flexbox h-full and stretch to keep columns aligned

Further simplify right column: keep only multi-label cards and new analysis button to balance UI layout

Update decision logic: compare calibrated probabilities (relative confidence) of Normal and disease classes

Implement Clinical Mutual Exclusion and Probability Calibration for intuitive UI rendering

Replace 6 old folds with 4 new folds (fold 0-3) and restore ensemble inference

Switch classification model from multi-fold ensemble to single fold 0 checkpoint

Correct rotation angle: rotate 90 degrees counter-clockwise to point frontal lobe upwards

Rotate display image 90 degrees clockwise on frontend canvas to make brain upright for visual presentation

Simplify UI: remove PathologyFocus component and hide AI Focus marks from MriSequenceViewer

Fix Focus slices labels: return useful_idx from backend to map raw indices to actual source slice numbers

Fix Focus slices alignment: return raw and mapped slice indices to support both DICOM and NIfTI rendering

Fix NIfTI preprocessing logic: remove rot90 and tissue_ratio to align 100% with Kaggle training

Support displaying NIfTI brain slices using Base64 image strings from backend

Replace legacy mri-46 DICOM demo sequence with pre-stripped NIfTI file mri-46_stripped.nii

Support direct NIfTI file upload (.nii, .nii.gz) bypassing skull stripping

Upgrade to CASA-MIL architecture with Transformer and 6-fold checkpoints

Fix: add mri-46 directory to Dockerfile copy and upgrade path resolution logic in main.py

UI: remove header, center logo rectangle, integrate mri-46 demo folder tracked by Git LFS

UI Redesign: centered header, enlarged logo, clinical demo cases, and dicom canvas drawer

UI Redesign: beautiful bg-teal-600 header and UploadZone with light teal borders and dynamic shadow

UI: simplified bg-teal-950 header, keep only Thuyloi University logo

UI Redesign: clean premium loading radar animation and teal-emerald color theme

Fix PyTorch CPU mixed precision float16 crash

Fix Debian package name libgl1-mesa-glx to libgl1 in Dockerfile

Add Hugging Face Space metadata config in README.md

Deploy MRI Pathology Detection App with LFS for all binaries