Note: This repository is a copy of the original code hosted at gitlab.inria.fr/yassis/DeepAneDet, developed during my PhD at Inria/LORIA.
Detection of intracranial aneurysms in 3D Time-of-Flight MRA as an object detection problem: a 3D single-stage, anchor-free network predicts each aneurysm as a sphere. This is the code and the annotations of the paper "Intracranial Aneurysm Detection: An object detection perspective" (IJCARS 2024), packaged as a set of commands that go from scans to detections and evaluation metrics.
Each aneurysm is approximated by a sphere defined by two points. The network divides each 3D patch into a grid of cells; every cell predicts the confidence that an aneurysm is centered in it, the position of that center and the sphere's radius. The data strategy of DeepAneSeg tackles the class imbalance: fast two-point annotation, negative patches centered on vessels and in the parenchyma, and aneurysm patches duplicated and deformed. On 402 subjects, the method reached an average precision of 78.96%, a sensitivity of 86.78% and 0.53 false positives per case, comparable or superior to the SCPM-Net [1], nnDetection [2] and nnU-Net [3] baselines.
Abstract
Intracranial aneurysm detection from 3D Time-Of-Flight Magnetic Resonance Angiography images is a problem of increasing clinical importance. Recently, a streak of methods have shown promising performance by using segmentation neural networks. However, these methods may be less relevant in a clinical settings where diagnostic decisions rely on detecting objects rather than their segmentation. We introduce a 3D single-stage object detection method tailored for small object detection such as aneurysms. Our anchor-free method incorporates fast data annotation, adapted data sampling and generation to address class imbalance problem, and spherical representations for improved object detection. A comprehensive evaluation was conducted, comparing our method with the state-of-the-art SCPM-Net[1], nnDetection[2] and nnUNet[3] baselines, using two datasets comprising 402 subjects, including one public dataset [4]. The evaluation used adapted object detection metrics. Our method exhibited comparable or superior performance, with an average precision of 78.96%, sensitivity of 86.78%, and 0.53 false positives per case. Our method significantly reduces the detection complexity compared to existing methods, and highlights the advantages of object detection over segmentation-based approaches for aneurysm detection. It also holds potential for application to other small object detection problems.
Detected aneurysms (red spheres) and their confidence:
git clone https://github.com/youssefassis/DeepAneDet && cd DeepAneDet
uv sync --extra cu126 # or --extra cpu, see Installation
uv run deepanedet-remove-skull data_dir=Data # lay out your data as in Data
uv run deepanedet-extract-points data_dir=Data
uv run deepanedet-train data_dir=Data data.split_file=reproducibility/fold1.json name=fold1
uv run deepanedet-predict train_dir=Data/0Work/fold1
uv run deepanedet-evaluate train_dir=Data/0Work/fold1 predictions=Data/0Work/fold1/prediction/200_epochs
Each step is described in Usage, and its results in Outputs.
The project uses uv and Python 3.10+. From the repository folder, pick the PyTorch build:
uv sync --extra cu126 # NVIDIA GPU (CUDA 12.6)
uv sync --extra cpu # CPU only, also the build used on macOS
This creates .venv with the exact versions of uv.lock. Commands then run with uv run, or directly once .venv
is activated.
Without uv, pip install git+https://github.com/youssefassis/DeepAneDet installs the package and its commands
(with PyTorch's default build from PyPI).
Each patient is a folder, either P0001-like or sub-XXX/ses-YYYYMMDD as in the public dataset [4]:
Data/
P0001/
volume.nii.gz TOF-MRA scan
aneurysms.csv aneurysms: two points (x, y, z columns, in mm) per aneurysm, see Annotation
config.json {"init volume": "volume.nii.gz"}
noskull.nii.gz skull-stripped scan (written by deepanedet-remove-skull)
points.csv negative patch centers (written by deepanedet-extract-points)
sub-013/
ses-20101220/
...
0Work/ trainings and logs (see Outputs)
Patients without aneurysm have no aneurysms.csv. Coordinates are in the scanner (RAS) space of the NIfTI file. A
split file lists the training, validation and testing patient folders, relative to the data folder or absolute, as in
reproducibility/fold1.json.
Each aneurysm is approximated by a sphere defined by two points: the center of its neck (F1) and its dome (F2).
1. Preprocessing: skull-strip the scans, then select the negative patch centers.
uv run deepanedet-remove-skull data_dir=Data
uv run deepanedet-extract-points data_dir=Data
2. Training: on the training patients, with the validation loss computed after each epoch; outputs go to
Data/0Work/fold1, and running the same command again resumes the training.
uv run deepanedet-train data_dir=Data data.split_file=reproducibility/fold1.json name=fold1
3. Prediction: writes the detections of each validation and testing patient, with the last checkpoint by
default, to prediction/<epochs>_epochs. Each detection has a center (x, y, z) and a radius in mm, and a confidence
between 0 and 1.
uv run deepanedet-predict train_dir=Data/0Work/fold1 # checkpoint=best_checkpoint.pytorch tta=false
4. Evaluation: matches the detections, from the most to the least confident, to the aneurysm they overlap most (IoU ≥ 0.1), and writes a CSV table of the detections and a summary figure next to the predictions.
uv run deepanedet-evaluate train_dir=Data/0Work/fold1 predictions=Data/0Work/fold1/prediction/200_epochs
Settings are Hydra configs in deepanedet/configs/: one file per command, composed from groups
(data, model, optimizer, scheduler, augmentation). Any value can be overridden on the command line:
uv run deepanedet-train data_dir=Data data.split_file=split.json name=exp2 batch_size=8 optimizer.lr=5e-3
uv run deepanedet-train --cfg job --resolve data_dir=Data data.split_file=split.json name=exp2 # print the config
To add an alternative, e.g. another optimizer, add a file with a _target_ class to the group, such as
deepanedet/configs/optimizer/adam.yaml, and select it with optimizer=adam.
Runs are reproducible: seed (0 by default) fixes the training, and each training records its git commit and library
versions in run_info.json. For identical GPU runs, also set deterministic=true (slower).
Everything is written under Data/0Work:
logs/ logs of the preprocessing commands
fold1/ one training
.hydra/config.yaml resolved configuration, read back by deepanedet-predict and deepanedet-evaluate
run_info.json git commit and library versions
last_checkpoint.pytorch model after the last epoch, used for prediction by default
best_checkpoint.pytorch model with the lowest validation loss
<epoch>_checkpoint.pytorch model every save_every (10) epochs
logs/ TensorBoard logs: uv run tensorboard --logdir Data/0Work
train.log
prediction/<epochs>_epochs/ for each validation and testing patient:
<patient>.json detections (P0001.json, sub-013_ses-20101220.json)
detection_evaluation@0.1.csv matched detections, written by deepanedet-evaluate
Evaluation@0.1.png evaluation figure
- Memory: training keeps the training and validation scans in memory, 4 bytes per voxel: 150 MiB for a 512×512×150 scan.
- GPU: the network has 28.8 M parameters (110 MiB). Its memory use comes mostly from the batch of 32 patches of
96³ voxels, with mixed precision; if the GPU runs out of memory, lower
batch_size.
reproducibility/annotations: annotations of the public dataset [4], one JSON file per subject. Write them as theaneurysms.csvof each patient folder with:uv run deepanedet-convert-annotations annotations_dir=reproducibility/annotations data_dir=Datareproducibility/fold1.jsontofold5.json: the 5-fold cross-validation splits, one training each:uv run deepanedet-train data_dir=Data data.split_file=reproducibility/fold1.json name=fold1baselines: the nnU-Net [3] and nnDetection [2] baselines (see its README).
uv run pytest # tests
uv run ruff check . # lint
uv run ruff format . # formatting
The code is organised as follows:
deepanedet/
cli/ commands
configs/ Hydra configuration
data/ patient data I/O, patch sampling, augmentation transforms
volume/ patch extraction, sphere burning, point selection, skull stripping
models/ single-stage sphere detection network
training/ dataset, balanced batch sampler, losses, schedulers, trainer
inference/ patch-wise prediction, detection matching and evaluation
reproducibility/ annotations of the public dataset and cross-validation splits
baselines/ nnU-Net and nnDetection baselines of the paper
If you find this work useful, please cite the paper (also available from "Cite this repository" on GitHub):
@article{assis2024intracranial,
title={Intracranial aneurysm detection: an object detection perspective},
author={Assis, Youssef and Liao, Liang and Pierre, Fabien and Anxionnat, Ren{\'e} and Kerrien, Erwan},
journal={International Journal of Computer Assisted Radiology and Surgery},
volume={19},
number={9},
pages={1667--1675},
year={2024},
publisher={Springer},
doi={10.1007/s11548-024-03132-z}
}
[1] Xiangde et al., SCPM-Net: An anchor-free 3D lung nodule detection network using sphere representation and center points matching. Medical Image Analysis 75, 102287 (2022).
[2] Baumgartner et al., nnDetection: A self-configuring method for medical object detection. MICCAI, pp. 530–539 (2021).
[3] Isensee et al., nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods 18(2), pp. 203–211 (2021).
[4] Di Noto et al., Towards automated brain aneurysm detection in TOF-MRA: open data, weak labels, and anatomical knowledge. Neuroinformatics, pp. 1–14 (2022).
This work was funded by the Grand-Est Region and the University Hospital (CHRU) of Nancy, France.