Clarify output labels and training mixture

Explain aggregate spine annotations without presenting them as checkpoint channels, and direct readers to the paper for training mixture proportions.

README.md

@@ -141,11 +141,11 @@ resample back to the original image size.
 
 ## Output labels
 
-61 channels: `background` at index 0, then 60 anatomical structures in the
-order stored in `label_schema.json`. The training protocol
-(`all_structures_flexray_v4`) has 62 foreground labels; its two coarse region
-labels, `lumbar_spine` and `thoracolumbar_spine`, only harmonize datasets that
-label those regions as a block and are not predicted.
+FleXray outputs 60 foreground masks (61 channels including `background`). The
+broader dataset/evaluation protocol also recognizes aggregate `lumbar_spine` and
+`thoracolumbar_spine` annotations; these are evaluated by combining the relevant
+per-vertebra outputs and are not checkpoint channels. Channel order is stored in
+`label_schema.json`.
 
 - **Skull / shoulder girdle:** skull, scapulae, clavicles
 - **Upper limb:** humeri, radii, ulnae, carpals, metacarpals, phalanges
@@ -168,17 +168,16 @@ architecture is in `config.yml`.
 ## Training data
 
 The model was trained on three source types unified under the FleXray label
-protocol (mixing proportions in parentheses):
+protocol. For training mixture proportions, please refer to the paper:
 
-- **Real X-ray masks:** HandBones (0.01), FootBones (0.01), MURA forearm
-  (0.01) and MURA humerus (0.01) with our own annotations.
-- **Generated FluXray images (0.375):** digitally reconstructed radiographs from
-  the MOOSE CTs, generatively edited toward real X-ray appearance, with exact
+- **Real X-ray masks:** HandBones, FootBones, MURA forearm, and MURA humerus,
+  with our own annotations.
+- **Generated FluXray images:** digitally reconstructed radiographs from the
+  MOOSE CTs, generatively edited toward real X-ray appearance, with exact
   overlapping masks for every protocol structure.
-- **Online CT->DRR rendering:** MOOSE / ENHANCE-PET 1.6k (0.375), Shoulder-CT
-  (0.05), HaN-Seg (0.05), PedsCT (0.04), RSNA cervical-spine fracture CTs
-  (0.04), and ElbowCT (0.03), rendered to DRRs at random poses during training
-  with per-label attenuation jitter.
+- **Online CT->DRR rendering:** MOOSE / ENHANCE-PET 1.6k, Shoulder-CT, HaN-Seg,
+  PedsCT, RSNA cervical-spine fracture CTs, and ElbowCT, rendered to DRRs at
+  random poses during training with per-label attenuation jitter.
 
 Training used AdamW (lr 3e-4, cosine schedule), a Dice + binary cross-entropy
 loss routed per dataset (partially labeled sources ignore unlabeled channels),