🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Check the NIfTI reader dependency, not just monai.

An existing MONAI installation without nibabel skips this install, but the later LoadImaged call reads downloaded .nii.gz files. Check import monai, nibabel (and install through the kernel interpreter) before continuing.

-!python -c "import monai" || pip install -q "monai-weekly[nibabel, tqdm]"
+!python -c "import monai, nibabel" || python -m pip install -q "monai-weekly[nibabel, tqdm]"

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@modules/rankseg_integration.ipynb` at line 47, Update the dependency-check
command in the notebook setup cell to validate both MONAI and nibabel imports
before skipping installation. Ensure any installation runs through the active
kernel interpreter, and retain the existing MONAI extras including nibabel and
tqdm so the later LoadImaged call can read .nii.gz files.

An alternative is to use %pip install -q --upgrade-strategy only-if-needed "monai-weekly[nibabel, tqdm,matplotlib,huggingface_hub]" rankseg in place of the multiple pip commands. This won't install over top of existing MONAI installs but will install missing dependencies.

#!/bin/bash
set -euo pipefail

python - <<'PY'
import json
from pathlib import Path

path = Path("modules/rankseg_integration.ipynb")
notebook = json.loads(path.read_text())
for index, cell in enumerate(notebook["cells"]):
    if cell.get("cell_type") != "code":
        continue
    source = "".join(cell.get("source", []))
    if any(term in source for term in (
        "pip install", "%pip", "import ", "from ", "rankseg", "huggingface_hub"
    )):
        print(f"\n--- code cell {index} ---")
        print(source)
PY

The PR description image shows six panels: a grayscale “CT image”; a white-mask “Tumor reference”; a purple/red/yellow “Tumor probability” heatmap; “Argmax tumor Dice 0.673”; “RankSEG tumor Dice 0.811”; and a “Changed tumor voxels” difference outline. The RankSEG result visibly aligns more closely with the reference mask.

@ericspod, agreed. This is a cleaner solution for the current setup cell: it replaces all four conditional installs, runs through the active notebook kernel, and declares every dependency the notebook imports—including nibabel for LoadImaged.

%pip install -q --upgrade-strategy only-if-needed "monai-weekly[nibabel, tqdm,matplotlib,huggingface_hub]" rankseg

It is a valid alternative to the import-guard change I suggested, so no separate import monai, nibabel check is needed if this consolidated command replaces the existing install commands.

Learnt from: ericspod
Repo: Project-MONAI/tutorials PR: 2068
File: modules/rankseg_integration.ipynb:47-47
Timestamp: 2026-07-27T12:45:08.827Z
Learning: In `modules/rankseg_integration.ipynb`, prefer one kernel-aware `%pip install -q --upgrade-strategy only-if-needed "monai-weekly[nibabel, tqdm,matplotlib,huggingface_hub]" rankseg` setup command over separate guarded `pip` commands. It covers the notebook’s MONAI NIfTI reader, tqdm, Matplotlib, Hugging Face Hub, and RankSEG runtime dependencies without unnecessarily replacing a compatible existing MONAI installation.

It's possible to use the bundle parser to load the model for you:

p = monai.bundle.ConfigParser()
p.read_config(bundle_dir + "/configs/inference.yaml")
p["arch_ckpt"] = architecture
model = p.get_parsed_content("network_def")

An alternative is to use %pip install -q --upgrade-strategy only-if-needed "monai-weekly[nibabel, tqdm,matplotlib,huggingface_hub]" rankseg in place of the multiple pip commands. This won't install over top of existing MONAI installs but will install missing dependencies.