Archive final baseline evaluation artifacts

baseline_final_evaluation_summary.json

@@ -0,0 +1,114 @@
+{
+    "archive_metadata": {
+        "archive_name": "Baseline 3D U-Net Final Evaluation Reference",
+        "created_utc": "2026-09-02T17:27:37.247327+00:00",
+        "evaluation_block": "Block 13",
+        "archive_block": "Block 14",
+        "status": "FINAL_BASELINE_REFERENCE"
+    },
+    "repository": {
+        "huggingface_repo": "Reduanul1997/RepositoryBaseLine3DuNet",
+        "repo_type": "model"
+    },
+    "model": {
+        "name": "Original3DUNet",
+        "architecture": "Baseline 3D U-Net",
+        "input_channels": 4,
+        "output_classes": 4,
+        "parameters": 22580964
+    },
+    "dataset": {
+        "dataset": "BraTS2020",
+        "modalities": [
+            "FLAIR",
+            "T1",
+            "T1ce",
+            "T2"
+        ],
+        "classes": [
+            "Background",
+            "NCR/NET",
+            "Edema",
+            "Enhancing Tumor"
+        ],
+        "training_patients": 294,
+        "validation_patients": 74,
+        "patch_size": [
+            64,
+            64,
+            64
+        ]
+    },
+    "model_selection": {
+        "best_model_epoch": 45,
+        "best_validation_dice_from_training": 0.772125,
+        "note": "This is the validation Dice reported by the training pipeline and used for model selection. It is distinct from the independently recomputed full-validation confusion-matrix metrics."
+    },
+    "evaluation_reference": {
+        "foreground_macro": {
+            "dice": 0.757824,
+            "iou": 0.618727,
+            "precision": 0.781685,
+            "recall": 0.740964
+        },
+        "tumor_binary": {
+            "dice": 0.927715,
+            "iou": 0.865175,
+            "precision": 0.942361,
+            "recall": 0.913516,
+            "sensitivity": 0.913516,
+            "false_negative_rate": 0.086484,
+            "false_positive_rate": 0.017968,
+            "TP": 4312069,
+            "TN": 14414613,
+            "FP": 263746,
+            "FN": 408228
+        },
+        "per_class": {
+            "Unknown": {
+                "Model": "Original3DUNet",
+                "Best_Model_Epoch": 45.0,
+                "Class_ID": 3.0,
+                "Class_Name": "Enhancing Tumor",
+                "Dice": 0.8512525465317852,
+                "IoU": 0.7410267104068393,
+                "Precision": 0.8501443117544676,
+                "Recall": 0.8523636744354415,
+                "Sensitivity": 0.8523636744354415,
+                "FNR": 0.1476363255645499,
+                "FPR": 0.0096019973023124,
+                "TP": 993227.0,
+                "TN": 18058317.0,
+                "FP": 175077.0,
+                "FN": 172035.0
+            }
+        }
+    },
+    "baseline_comparison_policy": {
+        "primary_segmentation_metric": "Foreground Macro Dice",
+        "secondary_segmentation_metrics": [
+            "Foreground Macro IoU",
+            "Foreground Macro Precision",
+            "Foreground Macro Recall"
+        ],
+        "primary_fn_metric": "Tumor False Negative Rate (FNR)",
+        "additional_fn_metric": "Tumor False Negatives (FN)",
+        "additional_tumor_metrics": [
+            "Tumor Dice",
+            "Tumor IoU",
+            "Tumor Precision",
+            "Tumor Recall",
+            "Tumor Sensitivity",
+            "Tumor FPR"
+        ],
+        "comparison_rule": "Future models must be evaluated using the same validation split, preprocessing, class definitions, tumor-vs-background definition, and metric calculation procedure before direct comparison with this baseline."
+    },
+    "artifact_files": [
+        "baseline_evaluation_results.json",
+        "baseline_evaluation_metrics.csv",
+        "baseline_per_class_metrics.csv",
+        "baseline_multiclass_confusion_matrix.png",
+        "baseline_tumor_background_confusion_matrix.png",
+        "baseline_final_evaluation_summary.json"
+    ]
+}