seqseg 2.1.7__tar.gz → 2.1.9__tar.gz

{seqseg-2.1.7 → seqseg-2.1.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: seqseg
-Version: 2.1.7
+Version: 2.1.9
 Summary: A deep learning-based medical image blood vessel tracking and segmentation tool.
 Author-email: Numi Sveinsson Cepero <numi@berkeley.com>
 License: Apache-2.0
@@ -30,8 +30,8 @@ Provides-Extra: viz
 Requires-Dist: matplotlib>=3.5; extra == "viz"
 Dynamic: license-file
 
-![example workflow](https://github.com/numisveinsson/SeqSeg/actions/workflows/python-app.yml/badge.svg)
-![example workflow](https://github.com/numisveinsson/SeqSeg/actions/workflows/test.yml/badge.svg)
+![test seqseg without torch build](https://github.com/numisveinsson/SeqSeg/actions/workflows/python-app.yml/badge.svg)
+![test seqseg build](https://github.com/numisveinsson/SeqSeg/actions/workflows/test.yml/badge.svg)
 
 <p align="center">
   <img src="https://raw.githubusercontent.com/numisveinsson/SeqSeg/main/seqseg/assets/seqseg_logo.png" alt="SeqSeg — Sequential Vessel Segmentation" width="480"/><br/>
@@ -50,10 +50,6 @@ Dynamic: license-file
   <a href="https://www.python.org"><img src="https://img.shields.io/badge/Python-3.9%2B-blue.svg" alt="Python 3.9+"/></a>
 </p>
 
-> **News:** SeqSeg writes a full SimVascular project in the `simvascular/` subdirectory — open it directly in SimVascular with pathlines and contours for every segmented branch.
-
----
-
 ## Why SeqSeg?
 
 SeqSeg segments vessels **sequentially**, taking steps along vessel centerlines and detecting bifurcations to grow complete vascular trees from just **1–2 seed points**. By combining local deep-learning predictions (nnU-Net) with geometric tracking, it stays robust across vessel scales — from small coronaries to large aortas.
@@ -65,57 +61,38 @@ SeqSeg segments vessels **sequentially**, taking steps along vessel centerlines
 - ✅ **Clinically validated** — coronary, aortic, cerebral, and pulmonary anatomies (pre-trained weights for aorta CT/MR and coronary CT)
 - ⚡ **Fast** — on CPU, about 2–5 minutes for an aorta, 5–15 for a coronary tree, and 10–30 for a cerebral tree. GPU is faster. Details in [Benchmarks](docs/benchmarks.md).
 
+A finished case includes a `simvascular/` project you can open directly in SimVascular, with pathlines and contours for every segmented branch.
+
 <p align="center">
   <img src="https://raw.githubusercontent.com/numisveinsson/SeqSeg/main/seqseg/assets/mr_model_tracing_fast_shorter.gif" alt="SeqSeg Demo"/><br/>
   <i>Real-time demonstration: automatic segmentation of an abdominal aorta in a 3D MR scan.</i>
 </p>
 
-## What's new in 2.x
-
-SeqSeg **2.0** uses subcommands. A command from 1.x that starts with a flag, such as `seqseg -data_dir ...`, still runs: it is rewritten to `seqseg run batch` automatically.
-
-### Commands
-
-| Command | Description |
-| -------- | ----------- |
-| **`seqseg run batch`** | Dataset batch tracing |
-| **`seqseg run single`** | One volume and seeds. Stages files under `<outdir>/_seqseg_single_staging/`, then traces |
-| **`seqseg run plus batch`** | Global nnU-Net sweep, then SeqSeg |
-| **`seqseg init dataset`** | Create `images/`, `centerlines/`, `truths/`, and a template `seeds.json` |
-| **`seqseg paths init` / `set` / `show`** | Save default nnU-Net, data, and output directories in `~/.seqseg/paths.yaml` |
-| **`seqseg train prepare`** | Extract patches or whole volumes and build an nnU-Net dataset |
-| **`seqseg train nnunet`** | Run nnU-Net planning, preprocessing, and training |
-| **`seqseg doctor`** | Check imports (SimpleITK, vtk, nnunetv2, scipy, vascular-segment-sampler) and paths |
-| **`seqseg config dump` / `fingerprint`** | Print a packaged YAML config, or list keys that differ from another |
-| **`seqseg post global-centerline`** | Build global centerlines from existing segmentations |
-| **`seqseg simvascular init`** | Create or refresh a SimVascular project under a case directory |
-| **`seqseg --version`** | Print the installed version |
-
-`seqseg run batch` flags use underscores (`-outdir` / `--outdir`). The data folder is `-data_dir` or `--data_directory`. `seqseg run single`, `seqseg train`, and `seqseg paths` use hyphens (`--image`, `--data-dir`).
+## Start here
 
-### Migrating from 1.x
-
-1. Prefer `seqseg run batch`. The old flag-only form still works.
-2. Use `seqseg run plus batch` in place of `python -m seqseg.seqseg_plus`. The nnU-Net path flags are unchanged.
-3. Check the install with `seqseg --version` after `pip install -U seqseg`.
+- **First segmentation** — the [tutorial](seqseg/tutorial/tutorial.md) runs a sample abdominal-aorta MR scan.
+- **Your own images** — [Usage](docs/usage.md) describes `images/` and `seeds.json`. The command is in [Quick Start](#quick-start).
+- **Train a model** — [Training](docs/training.md).
 
 ## Quick Start
 
 Python 3.9 or newer. 3.11 is the version used in the tutorial and conda example.
 
+The [tutorial](seqseg/tutorial/tutorial.md) is the sample run (abdominal-aorta MR, seed points, and Windows notes). Plan on 15–20 minutes including setup. Tracing is capped, so that part finishes in a few minutes; a real case should use the defaults.
+
+Install SeqSeg and the aorta weights below, then run `seqseg run batch` on your own data. The folder needs `images/` and `seeds.json` ([Usage](docs/usage.md)). This example is an aorta MR case. Change `-train_dataset`, `-config_name`, and `-img_ext` to match your images and weights.
+
 ```bash
 pip install seqseg
 
-# Aorta and femoral weights (MR and CT). Coronary weights are a separate zip; see Installation.
+# Aorta and femoral weights (MR and CT)
 curl -L -o nnUNet_results.zip https://zenodo.org/records/15020477/files/nnUNet_results.zip
 unzip nnUNet_results.zip
 ```
 
-`-nnunet_results_path` is the folder that contains `Dataset005_SEQAORTANDFEMOMR` (and the other dataset folders in that zip).
-
-**First run:** follow the [step-by-step tutorial](seqseg/tutorial/tutorial.md). It includes an abdominal-aorta MR scan, seed points, and Windows notes. The tutorial caps the number of steps so the example finishes in a few minutes; a real case should use the defaults.
+[Aorta and femoral weights](https://zenodo.org/records/15020477) (`nnUNet_results.zip`). [Coronary CT weights](https://zenodo.org/records/19547894) are a separate archive, `nnUNet_results_coronary.zip`.
 
-**Your own data** needs `images/` and `seeds.json` (see [Usage](docs/usage.md)). This command is an aorta MR case. Change `-train_dataset`, `-config_name`, and `-img_ext` to match your images and weights:
+`-nnunet_results_path` is the folder that contains `Dataset005_SEQAORTANDFEMOMR` (and the other dataset folders in that zip).
 
 ```bash
 seqseg run batch \
@@ -133,21 +110,21 @@ seqseg run batch \
 | Aorta / femoral CT | `Dataset006_SEQAORTANDFEMOCT` | `global_aorta` |
 | Coronary CT | `Dataset010_SEQCOROASOCACT` | `global_coro` |
 
-The tutorial uses `-config_name aorta_tutorial` for its sample scan. Download links and the trainer-folder layout are in [Installation](docs/installation.md).
+The tutorial uses `-config_name aorta_tutorial` for its sample scan. The trainer-folder layout is in [Installation](docs/installation.md). The full command list and flag spelling are in [Usage](docs/usage.md).
 
 ## Documentation
 
-| Guide | Description |
-|-------|-------------|
-| [Installation](docs/installation.md) | Setup, dependencies, and pre-trained model weights |
-| [Tutorial](seqseg/tutorial/tutorial.md) | End-to-end aorta example with sample data |
-| [Usage](docs/usage.md) | Data layout, seeds, CLI arguments, and output files |
-| [Configuration](docs/configuration.md) | YAML configs and tracking parameters |
-| [Algorithm Overview](docs/algorithm.md) | Methodology, workflow, and training strategy |
-| [Training](docs/training.md) | Train nnU-Net models on a new dataset |
-| [Performance & Benchmarks](docs/benchmarks.md) | Accuracy, timing, and qualitative comparisons |
-| [API](docs/api.md) | Call tracing from Python |
-| [Research & Development](docs/development.md) | SimVascular, 3D Slicer, and repository layout |
+| Job | Guide | Description |
+|-----|-------|-------------|
+| Run | [Installation](docs/installation.md) | Setup, dependencies, and pre-trained model weights |
+| Run | [Tutorial](seqseg/tutorial/tutorial.md) | End-to-end aorta example with sample data |
+| Run | [Usage](docs/usage.md) | Data layout, seeds, CLI arguments, and output files |
+| Run | [Configuration](docs/configuration.md) | YAML configs and tracking parameters |
+| Train | [Training](docs/training.md) | Train nnU-Net models on a new dataset |
+| Understand | [Algorithm Overview](docs/algorithm.md) | Methodology, workflow, and training strategy |
+| Understand | [Performance & Benchmarks](docs/benchmarks.md) | Accuracy, timing, and qualitative comparisons |
+| Integrate | [API](docs/api.md) | Call tracing from Python |
+| Integrate | [Research & Development](docs/development.md) | SimVascular, 3D Slicer, and repository layout |
 
 ## Citation
 

seqseg-2.1.9/README.md

@@ -0,0 +1,114 @@
+![test seqseg without torch build](https://github.com/numisveinsson/SeqSeg/actions/workflows/python-app.yml/badge.svg)
+![test seqseg build](https://github.com/numisveinsson/SeqSeg/actions/workflows/test.yml/badge.svg)
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/numisveinsson/SeqSeg/main/seqseg/assets/seqseg_logo.png" alt="SeqSeg — Sequential Vessel Segmentation" width="480"/><br/>
+  <img src="https://raw.githubusercontent.com/numisveinsson/SeqSeg/main/seqseg/assets/coronary.png" alt="Example coronary segmentation (SeqSeg)" width="260"/>
+</p>
+
+<h1 align="center">SeqSeg: Sequential Vessel Segmentation and Tracking</h1>
+
+<p align="center">
+  <b>Automatic tracking and segmentation of blood vessels in CT and MR images using deep learning and geometric tracking.</b>
+</p>
+
+<p align="center">
+  <a href="https://rdcu.be/dU0wy"><img src="https://img.shields.io/badge/Paper-Annals%20of%20BME-blue" alt="Paper"/></a>
+  <a href="https://github.com/numisveinsson/SeqSeg/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-Apache%202.0-green.svg" alt="License"/></a>
+  <a href="https://www.python.org"><img src="https://img.shields.io/badge/Python-3.9%2B-blue.svg" alt="Python 3.9+"/></a>
+</p>
+
+## Why SeqSeg?
+
+SeqSeg segments vessels **sequentially**, taking steps along vessel centerlines and detecting bifurcations to grow complete vascular trees from just **1–2 seed points**. By combining local deep-learning predictions (nnU-Net) with geometric tracking, it stays robust across vessel scales — from small coronaries to large aortas.
+
+- 🌱 **Minimal supervision** — only 1–2 seed points to initialize
+- 🌿 **Robust bifurcation detection** — automatically follows every branch
+- 🩻 **Multi-modal** — works with CT and MR 3D medical images
+- 📏 **Scalable** — vessels from ~1mm coronaries to ~30mm aortas
+- ✅ **Clinically validated** — coronary, aortic, cerebral, and pulmonary anatomies (pre-trained weights for aorta CT/MR and coronary CT)
+- ⚡ **Fast** — on CPU, about 2–5 minutes for an aorta, 5–15 for a coronary tree, and 10–30 for a cerebral tree. GPU is faster. Details in [Benchmarks](docs/benchmarks.md).
+
+A finished case includes a `simvascular/` project you can open directly in SimVascular, with pathlines and contours for every segmented branch.
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/numisveinsson/SeqSeg/main/seqseg/assets/mr_model_tracing_fast_shorter.gif" alt="SeqSeg Demo"/><br/>
+  <i>Real-time demonstration: automatic segmentation of an abdominal aorta in a 3D MR scan.</i>
+</p>
+
+## Start here
+
+- **First segmentation** — the [tutorial](seqseg/tutorial/tutorial.md) runs a sample abdominal-aorta MR scan.
+- **Your own images** — [Usage](docs/usage.md) describes `images/` and `seeds.json`. The command is in [Quick Start](#quick-start).
+- **Train a model** — [Training](docs/training.md).
+
+## Quick Start
+
+Python 3.9 or newer. 3.11 is the version used in the tutorial and conda example.
+
+The [tutorial](seqseg/tutorial/tutorial.md) is the sample run (abdominal-aorta MR, seed points, and Windows notes). Plan on 15–20 minutes including setup. Tracing is capped, so that part finishes in a few minutes; a real case should use the defaults.
+
+Install SeqSeg and the aorta weights below, then run `seqseg run batch` on your own data. The folder needs `images/` and `seeds.json` ([Usage](docs/usage.md)). This example is an aorta MR case. Change `-train_dataset`, `-config_name`, and `-img_ext` to match your images and weights.
+
+```bash
+pip install seqseg
+
+# Aorta and femoral weights (MR and CT)
+curl -L -o nnUNet_results.zip https://zenodo.org/records/15020477/files/nnUNet_results.zip
+unzip nnUNet_results.zip
+```
+
+[Aorta and femoral weights](https://zenodo.org/records/15020477) (`nnUNet_results.zip`). [Coronary CT weights](https://zenodo.org/records/19547894) are a separate archive, `nnUNet_results_coronary.zip`.
+
+`-nnunet_results_path` is the folder that contains `Dataset005_SEQAORTANDFEMOMR` (and the other dataset folders in that zip).
+
+```bash
+seqseg run batch \
+    -data_dir your_data/ \
+    -nnunet_results_path nnUNet_results/ \
+    -train_dataset Dataset005_SEQAORTANDFEMOMR \
+    -config_name global_aorta \
+    -img_ext .mha \
+    -outdir results/
+```
+
+| Anatomy | `-train_dataset` | `-config_name` |
+|---------|------------------|----------------|
+| Aorta / femoral MR | `Dataset005_SEQAORTANDFEMOMR` | `global_aorta` |
+| Aorta / femoral CT | `Dataset006_SEQAORTANDFEMOCT` | `global_aorta` |
+| Coronary CT | `Dataset010_SEQCOROASOCACT` | `global_coro` |
+
+The tutorial uses `-config_name aorta_tutorial` for its sample scan. The trainer-folder layout is in [Installation](docs/installation.md). The full command list and flag spelling are in [Usage](docs/usage.md).
+
+## Documentation
+
+| Job | Guide | Description |
+|-----|-------|-------------|
+| Run | [Installation](docs/installation.md) | Setup, dependencies, and pre-trained model weights |
+| Run | [Tutorial](seqseg/tutorial/tutorial.md) | End-to-end aorta example with sample data |
+| Run | [Usage](docs/usage.md) | Data layout, seeds, CLI arguments, and output files |
+| Run | [Configuration](docs/configuration.md) | YAML configs and tracking parameters |
+| Train | [Training](docs/training.md) | Train nnU-Net models on a new dataset |
+| Understand | [Algorithm Overview](docs/algorithm.md) | Methodology, workflow, and training strategy |
+| Understand | [Performance & Benchmarks](docs/benchmarks.md) | Accuracy, timing, and qualitative comparisons |
+| Integrate | [API](docs/api.md) | Call tracing from Python |
+| Integrate | [Research & Development](docs/development.md) | SimVascular, 3D Slicer, and repository layout |
+
+## Citation
+
+When using SeqSeg, please cite the following [paper](https://rdcu.be/dU0wy):
+
+```
+@Article{SveinssonCepero2024,
+author={Sveinsson Cepero, Numi
+and Shadden, Shawn C.},
+title={SeqSeg: Learning Local Segments for Automatic Vascular Model Construction},
+journal={Annals of Biomedical Engineering},
+year={2024},
+month={Sep},
+day={18},
+issn={1573-9686},
+doi={10.1007/s10439-024-03611-z},
+url={https://doi.org/10.1007/s10439-024-03611-z},
+}
+```

{seqseg-2.1.7 → seqseg-2.1.9}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "seqseg"
-version = "2.1.7"
+version = "2.1.9"
 description = "A deep learning-based medical image blood vessel tracking and segmentation tool."
 readme = "README.md"
 requires-python = ">=3.9"

{seqseg-2.1.7 → seqseg-2.1.9}/seqseg/cli.py

@@ -437,7 +437,7 @@ def _cmd_init_dataset(ns: argparse.Namespace) -> None:
         "-nnunet_results_path ...\n"
         "\nNext (train a new model): also fill centerlines/ and truths/, then:\n"
         f"  seqseg paths init --data-dir {root}/ --outdir ~/seqseg_train\n"
-        "  seqseg train prepare --name MYDATA --dataset-number 999 --modality CT\n"
+        "  seqseg train prepare --name MYDATA --dataset-id 999 --modality CT\n"
     )
 
 
@@ -634,6 +634,12 @@ def _cmd_paths_export(_ns: argparse.Namespace) -> None:
 
 
 def _cmd_train_prepare(ns: argparse.Namespace) -> None:
+    if not ns.preprocess_only and not ns.name:
+        print(
+            "seqseg train prepare: --name is required",
+            file=sys.stderr,
+        )
+        sys.exit(2)
     try:
         result = prepare_training_dataset(
             data_dir=ns.data_dir,
@@ -656,11 +662,14 @@ def _cmd_train_prepare(ns: argparse.Namespace) -> None:
             skip_sample=ns.skip_sample,
             skip_convert=ns.skip_convert,
             also_test=ns.also_test,
-            yes=ns.yes,
             verbose=ns.verbose,
             img_ext=ns.img_ext,
             global_volumes=ns.global_volumes,
             keep_extracted=ns.keep_extracted,
+            configuration=ns.configuration,
+            plans=ns.plans,
+            np=ns.np,
+            preprocess_only=ns.preprocess_only,
         )
     except TrainDependencyError as e:
         print(str(e), file=sys.stderr)
@@ -670,6 +679,9 @@ def _cmd_train_prepare(ns: argparse.Namespace) -> None:
         if message.startswith("Sampler config"):
             print(message, file=sys.stderr)
             sys.exit(2)
+        if "seqseg train prepare" in message:
+            print(message, file=sys.stderr)
+            sys.exit(1)
         import traceback
 
         traceback.print_exc()
@@ -691,17 +703,30 @@ def _cmd_train_prepare(ns: argparse.Namespace) -> None:
         sys.exit(1)
 
     print("\nPrepare complete.")
-    if result.removed_staging:
+    if ns.preprocess_only:
         print(
-            "  Extracted volumes were linked into nnU-Net raw and the "
-            "temporary folder was removed."
+            f"  Preprocessed {ns.configuration} from the existing raw dataset."
+        )
+    elif result.removed_staging:
+        print(
+            "  Extracted subvolumes were removed after conversion. "
+            "done.txt and the .txt/.csv metadata were kept in\n"
+            f"  {result.extracted_dir}"
         )
     else:
         kind = "Whole volumes" if ns.global_volumes else "Patches"
         print(f"  {kind}: {result.extracted_dir}")
+    if not result.preprocessed:
+        print(
+            "  nnU-Net preprocessing was skipped (--skip-convert). "
+            "Re-run prepare without --skip-convert before training."
+        )
     results = resolve_path("nnunet_results")
     if results:
         print(f"  Trained models will be written under: {results}")
+    config_flag = ""
+    if ns.configuration != "3d_fullres":
+        config_flag = f" --configuration {ns.configuration}"
     for name, path in zip(result.dataset_names, result.dataset_dirs):
         try:
             ds_id = dataset_id_from_name(name)
@@ -709,11 +734,12 @@ def _cmd_train_prepare(ns: argparse.Namespace) -> None:
             ds_id = "?"
         print(f"  nnU-Net dataset: {name}")
         print(f"    path: {path}")
-        print(
-            "    next:\n"
-            f"      seqseg train nnunet --dataset-id {ds_id} "
-            "--configuration 3d_fullres --fold 0"
-        )
+        if result.preprocessed:
+            print(
+                "    next:\n"
+                f"      seqseg train nnunet --dataset-id {ds_id}"
+                f"{config_flag} --fold 0"
+            )
 
 
 def _cmd_train_nnunet(ns: argparse.Namespace) -> None:
@@ -736,12 +762,10 @@ def _cmd_train_nnunet(ns: argparse.Namespace) -> None:
             dataset_id,
             configuration=ns.configuration,
             fold=ns.fold,
-            skip_plan=ns.skip_plan,
-            plan_only=ns.plan_only,
-            np=ns.np,
             trainer=ns.trainer,
             plans=ns.plans,
             cleanup=ns.cleanup,
+            continue_training=ns.continue_training,
         )
     except Exception as e:  # noqa: BLE001
         print(f"seqseg train nnunet failed: {e}", file=sys.stderr)
@@ -1021,7 +1045,7 @@ def _build_parser() -> argparse.ArgumentParser:
 
     p_prep = train_sub.add_parser(
         "prepare",
-        help="Prepare patches or whole volumes and convert to nnU-Net DatasetXXX format",
+        help="Extract volumes, write an nnU-Net dataset, and plan/preprocess",
         formatter_class=argparse.ArgumentDefaultsHelpFormatter,
     )
     p_prep.add_argument(
@@ -1038,22 +1062,27 @@ def _build_parser() -> argparse.ArgumentParser:
         default=None,
         type=str,
         help="Optional scratch directory for extracted patches or whole volumes. "
-        "Default: <parent of nnUNet_raw>/_seqseg_extracted, removed after "
-        "conversion. Pass this to keep extracts or resume with --skip-sample.",
+        "Default: nnUNet_raw/DatasetXXX_*/_seqseg_extracted. Subvolumes are "
+        "removed after conversion; done.txt and .txt/.csv metadata are kept. "
+        "Pass this to keep a copy elsewhere.",
     )
     p_prep.add_argument(
         "--name",
         "-name",
-        required=True,
+        default=None,
         type=str,
-        help="Dataset name stem (e.g. MYDATA → Dataset0XX_MYDATACT)",
+        help="Dataset name stem (e.g. MYDATA → Dataset0XX_MYDATACT). "
+        "Required unless --preprocess-only",
     )
     p_prep.add_argument(
+        "--dataset-id",
         "--dataset-number",
         "-dataset_number",
+        dest="dataset_number",
         required=True,
         type=int,
-        help="nnU-Net dataset id (e.g. 999). Incremented per modality if several",
+        help="nnU-Net dataset id to create (e.g. 999). "
+        "Incremented per modality if several",
     )
     p_prep.add_argument(
         "--modality",
@@ -1065,7 +1094,7 @@ def _build_parser() -> argparse.ArgumentParser:
     p_prep.add_argument(
         "--config-name",
         "-config_name",
-        default="global",
+        default="default_mm",
         type=str,
         help="Sampler YAML name/path (from vascular-segment-sampler)",
     )
@@ -1125,18 +1154,43 @@ def _build_parser() -> argparse.ArgumentParser:
         action="store_true",
         help="Also convert modality_test folders when present",
     )
-    p_prep.add_argument("--yes", action="store_true", help="Non-interactive confirms")
     p_prep.add_argument("--verbose", action="store_true")
     p_prep.add_argument(
         "--keep-extracted",
         action="store_true",
-        help="Keep <nnunet-root>/_seqseg_extracted after conversion into nnUNet_raw",
+        help="Keep extracted subvolumes in nnUNet_raw/DatasetXXX_*/_seqseg_extracted "
+        "after conversion into nnUNet_raw",
+    )
+    p_prep.add_argument(
+        "--configuration",
+        default="3d_fullres",
+        choices=["3d_fullres", "2d", "3d_lowres", "3d_cascade_fullres"],
+        help="nnU-Net configuration to preprocess. Default 3d_fullres",
+    )
+    p_prep.add_argument(
+        "--plans",
+        default="nnUNetPlans",
+        type=str,
+        help="nnU-Net plans name (training -p). Default nnUNetPlans",
+    )
+    p_prep.add_argument(
+        "--np",
+        type=int,
+        default=None,
+        help="Processes for nnU-Net preprocessing. Default: --num-cores",
+    )
+    p_prep.add_argument(
+        "--preprocess-only",
+        action="store_true",
+        help="Use an existing nnUNet_raw dataset. Plans and preprocesses it "
+        "when plans are missing; otherwise resamples --configuration. "
+        "Does not extract or convert subvolumes",
     )
     p_prep.set_defaults(_handler=_cmd_train_prepare, img_ext=None)
 
     p_nn = train_sub.add_parser(
         "nnunet",
-        help="Run nnUNetv2_plan_and_preprocess and/or nnUNetv2_train",
+        help="Train with nnUNetv2_train (--c continues from the latest checkpoint)",
         formatter_class=argparse.ArgumentDefaultsHelpFormatter,
     )
     p_nn.add_argument(
@@ -1149,13 +1203,13 @@ def _build_parser() -> argparse.ArgumentParser:
         "--dataset-name",
         type=str,
         default=None,
-        help="Alternative to --dataset-id (e.g. Dataset0999_MYDATACT)",
+        help="Alternative to --dataset-id (e.g. Dataset999_MYDATACT)",
     )
     p_nn.add_argument(
         "--configuration",
         default="3d_fullres",
         choices=["3d_fullres", "2d", "3d_lowres", "3d_cascade_fullres"],
-        help="nnU-Net configuration for preprocessing (-c) and training",
+        help="nnU-Net configuration to train. Default 3d_fullres",
     )
     p_nn.add_argument(
         "--fold",
@@ -1164,20 +1218,11 @@ def _build_parser() -> argparse.ArgumentParser:
         help="Fold to train (0-4 or all)",
     )
     p_nn.add_argument(
-        "--skip-plan",
-        action="store_true",
-        help="Skip plan_and_preprocess; only train",
-    )
-    p_nn.add_argument(
-        "--plan-only",
+        "--c",
+        "--continue",
+        dest="continue_training",
         action="store_true",
-        help="Only run plan_and_preprocess",
-    )
-    p_nn.add_argument(
-        "--np",
-        type=int,
-        default=None,
-        help="Processes for plan_and_preprocess (-np)",
+        help="Continue training from checkpoint_latest.pth (nnU-Net --c)",
     )
     p_nn.add_argument("--trainer", default="nnUNetTrainer", type=str)
     p_nn.add_argument("--plans", default="nnUNetPlans", type=str)
@@ -1186,7 +1231,7 @@ def _build_parser() -> argparse.ArgumentParser:
         action="store_true",
         help="After training succeeds, delete this dataset's nnUNet_raw and "
         "nnUNet_preprocessed folders. Leaves nnUNet_results. "
-        "Ignored with --plan-only.",
+        "A later --c run needs that preprocessed data.",
     )
     p_nn.set_defaults(_handler=_cmd_train_nnunet)
 

seqseg-2.1.9/seqseg/pipeline/nnunet_cli.py

@@ -0,0 +1,108 @@
+"""Run an nnU-Net planning or preprocessing entry point with a faster file lookup.
+
+``python -m seqseg.pipeline.nnunet_cli plan_and_preprocess -d 999 ...`` takes the
+same arguments as ``nnUNetv2_plan_and_preprocess``.
+
+nnU-Net finds a dataset's training files by matching every case id against
+every name in ``imagesTr``. That is quadratic in the number of cases, and it
+runs again in fingerprinting, planning, preprocessing, and the label copy:
+about an hour each for several hundred thousand patches. Listing the files
+under ``dataset`` in dataset.json avoids it, but nnU-Net sends dataset.json to
+every preprocessing job, so a long list slows preprocessing far more.
+
+This module swaps in a single-pass lookup before nnU-Net starts. It returns
+the same mapping as nnU-Net's folder scan. nnU-Net's files are not modified.
+"""
+
+from __future__ import annotations
+
+import os
+import re
+import sys
+from typing import Callable, Dict, List, Optional
+
+_ENTRY_POINTS = {
+    "plan_and_preprocess": "plan_and_preprocess_entry",
+    "preprocess": "preprocess_entry",
+}
+
+_CHANNEL_SUFFIX = re.compile(r"_\d{4}$")
+
+
+def training_files_from_folder(raw_dataset_folder: str, dataset_json: dict) -> Dict[str, dict]:
+    """Same result as nnU-Net's ``imagesTr`` scan, in one directory listing."""
+    ending = dataset_json["file_ending"]
+    crop = len(ending) + 5
+    images_dir = os.path.join(raw_dataset_folder, "imagesTr")
+    labels_dir = os.path.join(raw_dataset_folder, "labelsTr")
+
+    images: Dict[str, List[str]] = {}
+    with os.scandir(images_dir) as entries:
+        for entry in entries:
+            name = entry.name
+            if not name.endswith(ending) or not entry.is_file():
+                continue
+            files = images.setdefault(name[:-crop], [])
+            if _CHANNEL_SUFFIX.search(name[: -len(ending)]):
+                files.append(name)
+
+    return {
+        case: {
+            "images": [os.path.join(images_dir, name) for name in sorted(images[case])],
+            "label": os.path.join(labels_dir, case + ending),
+        }
+        for case in sorted(images)
+    }
+
+
+def _fast_lookup(original: Callable) -> Callable:
+    def get_filenames_of_train_images_and_targets(
+        raw_dataset_folder: str, dataset_json: Optional[dict] = None
+    ):
+        if dataset_json is None:
+            from batchgenerators.utilities.file_and_folder_operations import load_json
+
+            dataset_json = load_json(os.path.join(raw_dataset_folder, "dataset.json"))
+        if "dataset" in dataset_json:
+            return original(raw_dataset_folder, dataset_json)
+        return training_files_from_folder(raw_dataset_folder, dataset_json)
+
+    return get_filenames_of_train_images_and_targets
+
+
+def install_fast_lookup() -> None:
+    """Point every loaded nnU-Net module at the single-pass lookup.
+
+    Modules bind the function with ``from ... import``, so each one is patched.
+    Only the parent process looks up files; spawned workers do not need it.
+    """
+    import nnunetv2.experiment_planning.plan_and_preprocess_entrypoints  # noqa: F401
+    import nnunetv2.preprocessing.preprocessors.default_preprocessor  # noqa: F401
+    import nnunetv2.utilities.utils as nnunet_utils
+
+    original = nnunet_utils.get_filenames_of_train_images_and_targets
+    fast = _fast_lookup(original)
+    for name, module in list(sys.modules.items()):
+        if not name.startswith("nnunetv2") or module is None:
+            continue
+        if getattr(module, "get_filenames_of_train_images_and_targets", None) is original:
+            module.get_filenames_of_train_images_and_targets = fast
+
+
+def main(argv: Optional[List[str]] = None) -> None:
+    args = list(sys.argv[1:] if argv is None else argv)
+    if not args or args[0] not in _ENTRY_POINTS:
+        raise SystemExit(
+            "usage: python -m seqseg.pipeline.nnunet_cli "
+            f"{{{','.join(_ENTRY_POINTS)}}} [nnU-Net arguments]"
+        )
+    install_fast_lookup()
+    from nnunetv2.experiment_planning import plan_and_preprocess_entrypoints
+
+    command = args.pop(0)
+    sys.argv = [f"nnUNetv2_{command}", *args]
+    getattr(plan_and_preprocess_entrypoints, _ENTRY_POINTS[command])()
+
+
+if __name__ == "__main__":
+    main()