pith. sign in

arxiv: 2605.28217 · v1 · pith:DLS42JCRnew · submitted 2026-05-27 · 💻 cs.CV

A Patient-Specific Pulmonary Arterial Tree Digital Twin to Extract Pulmonary Embolism Biomarkers

Morgane des Ligneris , Nathan Painchaud , Allan Serva , Laurent Bertoletti , Pierre Croisille , Carole Frindel , Odyss\'ee Merveille This is my paper

Pith reviewed 2026-06-29 13:02 UTC · model grok-4.3

classification 💻 cs.CV
keywords pulmonary embolismdigital twinarterial treegraph representationseverity scoresQanadliMastoraCT pulmonary angiography
0
0 comments X

The pith

An automated pipeline converts AI-generated CT masks into a directed graph model of the pulmonary arterial tree to derive patient-specific embolism biomarkers and severity scores.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Pulmonary embolism diagnosis currently depends on risk stratification that often requires time-consuming manual scoring or blood tests not always available in emergencies. The paper presents a pipeline that takes binary masks of arteries, emboli, lungs, and lobes and constructs a labeled hierarchical graph of the arterial tree. From this graph the method extracts local features such as clot volume and obstruction at each artery segment plus global measures including automatically computed Qanadli and Mastora scores and embolic volume distributions by lobe and hierarchy level. Validation against an existing pipeline, anatomical expectations, and manual score calculations shows strong agreement, indicating the digital twin can supply rapid image-based information on clot burden and location.

Core claim

Using artificial-intelligence-generated binary masks as input, the pipeline builds a directed graph representation of the pulmonary arterial tree, labels its hierarchical structure, and extracts both artery-level biomarkers (morphology, position, clot volume, obstruction) and patient-level biomarkers (Qanadli and Mastora severity scores plus total embolic volume by lobes and levels), with the resulting digital twins and scores agreeing with manual calculations and anatomical expectations.

What carries the argument

The directed graph representation of the pulmonary arterial tree, constructed from binary masks, that labels hierarchical branches and enables extraction of local morphological and obstruction features plus global severity scores.

If this is right

  • Severity scores can be produced automatically without manual calculation or blood biomarkers.
  • Embolic burden can be quantified by lobe and by hierarchical level in the arterial tree.
  • Local obstruction and clot volume become available at every artery segment for finer spatial analysis.
  • The same graph model supports consistent comparison across patients or serial scans.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The graph structure could be reused to simulate flow changes after clot removal or to test alternative risk metrics.
  • Extending the same masking-to-graph approach to other vascular beds would require only retraining the upstream segmentation models.
  • Integration into emergency CT workflows could reduce the time from scan to risk stratification.

Load-bearing premise

The upstream AI-generated binary masks of arteries, emboli, lungs, and lobes must be sufficiently accurate for the graph construction and biomarker steps to produce reliable results.

What would settle it

A head-to-head comparison on a new cohort of CT pulmonary angiograms where the automated Qanadli and Mastora scores differ substantially from scores calculated manually by multiple radiologists.

Figures

Figures reproduced from arXiv: 2605.28217 by Allan Serva, Carole Frindel, Laurent Bertoletti, Morgane des Ligneris, Nathan Painchaud, Odyss\'ee Merveille, Pierre Croisille.

Figure 1
Figure 1. Figure 1: General workflow of the proposed pipeline. A) Input data: automatically segmented masks of arteries, PE, lungs, and lobes. B) Pipeline steps: (1) graph modeling from the arterial mask, (2) hierarchical labeling using lung and lobe masks, and (3) integration of obstruction-related information from the PE masks. Each step produces local biomarkers, which are subsequently aggregated in (4) patient-level bioma… view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Graph correction steps. A) Non-physiological cycle caused by a contact between two adjacent arteries in the segmentation mask, pro￾ducing an artificial local diameter constriction along a single artery seg￾ment, characterize by successive diameters dmax before, dmin, dmax after. B) Spurious terminal branch generated by surface irregularities of the segmentation mask. dparent denotes the diameter of the par… view at source ↗
Figure 4
Figure 4. Figure 4: Hierarchical levels of pulmonary arteries, each level is presented with a different color. as the MPA. Laterality is determined by intersecting each edge with the right and left lung masks. The first bifurcation distal to (after) the MPA defines the right and left pulmonary arteries (RPA and LPA, hierarchical level 2). Lobar affiliation is assigned by intersecting terminal arteries with the lobe masks and … view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of vascular graphs obtained on the same patient with a zoom view of the main pulmonary artery. (A) Graph generated by the Voreen VesselGraph pipeline, containing artifactual branches on the MPA. (B) Graph after our cleaning step, showing the removal of these artifacts [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example case of global biomarkers extracted to characterize the total embolic volume repartition by lobes and levels [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 1
Figure 1. Figure 1: Iterative training and configuration of nnU-Net segmentation models. (A) Iterative annotation and training protocol: a subset of patients is annotated, used to train nnU-Net models, predictions are reviewed and corrected, and the number of annotations is progressively increased until a final model is selected to segment the full database. (B) Anatomical contexts used to train the different nnU-Net models, … view at source ↗
read the original abstract

Pulmonary embolism, the obstruction of a pulmonary artery by a blood clot, is one of the leading causes of acute cardiovascular syndrome. In clinical practice, therapeutic decisions after diagnosis via computed tomography pulmonary angiography rely on risk stratification, which categorizes 30-day mortality risk into three categories. This stratification depends on the right-to-left ventricular diameter ratio and blood levels of two cardiac enzymes. However, blood biomarkers are not always available in emergency settings, and manual calculation of established severity scores - such as Qanadli and Mastora - is time-consuming and rarely performed in clinical routine practice. This study introduces an automated pipeline that models a directed graph representation of the pulmonary arterial tree, labeling its hierarchical structure and characterizing pulmonary embolism. The pipeline derives image-based biomarkers, including local artery-level features (morphological information, hierarchical position, clot volume, and resulting obstruction) and global patient-level biomarkers such as automatically calculated severity scores (Qanadli and Mastora) and the total embolic volume distribution by lobes and hierarchical levels. Using artificial-intelligence-generated binary masks of arteries, emboli, lungs, and lobes, it creates a patient digital twin of the arterial structure. Validation of the pipeline through comparison to an existing pipeline, anatomical expectations, and manual severity score calculations demonstrates the pipeline's ability to automatically generate anatomically accurate digital twins and severity scores with strong agreement. This supports the potential of these image-derived biomarkers to automatically provide rapid, precise information on thrombotic burden and spatial clot distribution.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper presents an automated pipeline that takes AI-generated binary masks of pulmonary arteries, emboli, lungs, and lobes as input, constructs a directed-graph representation of the patient-specific pulmonary arterial tree, performs hierarchical labeling, and extracts local artery-level biomarkers (morphology, position, clot volume, obstruction) together with global patient-level biomarkers including automatically computed Qanadli and Mastora severity scores and embolic volume distributions.

Significance. If the pipeline's outputs prove robust, the work would supply a practical route to rapid, image-derived risk stratification for pulmonary embolism that bypasses manual score calculation and blood-biomarker availability constraints, potentially improving routine clinical decision-making.

major comments (2)
  1. [Abstract] Abstract: the central claim of 'strong agreement' with manual severity scores and 'anatomically accurate digital twins' is asserted without any quantitative metrics, cohort size, exclusion criteria, or error statistics; the provided text supplies no numbers that would allow verification of the validation results.
  2. [Abstract / Validation] Validation description (abstract and implied results): the reported comparisons to an existing pipeline, anatomical expectations, and manual scores evaluate only downstream graph construction and obstruction calculations under the assumption that the upstream AI masks are already accurate; no sensitivity analysis, error-propagation study, or ground-truth segmentation comparison is described that would quantify how typical mask inaccuracies affect hierarchical labeling or final patient-level scores.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate revisions made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of 'strong agreement' with manual severity scores and 'anatomically accurate digital twins' is asserted without any quantitative metrics, cohort size, exclusion criteria, or error statistics; the provided text supplies no numbers that would allow verification of the validation results.

    Authors: We agree that the abstract, standing alone, does not contain the quantitative details needed to substantiate the claims. The full manuscript reports these metrics in the results, but to make the abstract verifiable we have revised it to include cohort size, specific agreement statistics with manual Qanadli/Mastora scores, and error measures. revision: yes

  2. Referee: [Abstract / Validation] Validation description (abstract and implied results): the reported comparisons to an existing pipeline, anatomical expectations, and manual scores evaluate only downstream graph construction and obstruction calculations under the assumption that the upstream AI masks are already accurate; no sensitivity analysis, error-propagation study, or ground-truth segmentation comparison is described that would quantify how typical mask inaccuracies affect hierarchical labeling or final patient-level scores.

    Authors: The validation targets the graph-construction and biomarker-extraction stages given the input masks; upstream segmentation performance is documented in the cited mask-generation studies. We acknowledge that explicit sensitivity testing would be valuable. We have therefore added a sensitivity-analysis subsection that perturbs the input masks according to reported segmentation error rates and quantifies downstream effects on labeling accuracy and patient-level scores. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The pipeline constructs a directed graph from upstream AI-generated binary masks of arteries, emboli, lungs, and lobes, then computes local artery-level features (morphology, hierarchical position, clot volume, obstruction) and global biomarkers (Qanadli/Mastora scores, embolic volume distribution) directly via graph traversal and volume calculations. These steps are algorithmic extractions from the input masks and graph structure, with no equations, fitted parameters, or self-referential definitions that reduce outputs to inputs by construction. Validation relies on comparison to an existing pipeline, anatomical expectations, and manual score calculations, which are independent of the automated steps. No self-citation chains or ansatzes are invoked as load-bearing for the core derivation. The method is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the accuracy of upstream AI segmentation masks and on the assumption that a directed graph faithfully captures the anatomical hierarchy needed for biomarker calculation.

axioms (1)
  • domain assumption AI-generated binary masks of arteries, emboli, lungs, and lobes are sufficiently accurate to support reliable graph construction and biomarker extraction.
    The abstract states that the pipeline uses these masks as direct input; any systematic error in the masks propagates to all downstream biomarkers.
invented entities (1)
  • Patient-specific pulmonary arterial tree digital twin no independent evidence
    purpose: Directed graph representation that encodes hierarchical structure, morphology, and obstruction for biomarker extraction.
    Introduced in the abstract as the core modeling step; no independent falsifiable prediction (e.g., a predicted mass or measurable property outside the imaging data) is supplied.

pith-pipeline@v0.9.1-grok · 5829 in / 1391 out tokens · 51299 ms · 2026-06-29T13:02:45.256412+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Pulmonary Embolism Risk Stratification from CTPA and Medical Records: Vascular Graphs Are Not All You Need

    cs.CV 2026-06 unverdicted novelty 5.0

    On a private dataset of 353 patients, medical records and cardiac biomarkers from CTPA outperform vascular biomarkers and GNNs on vascular graphs for PE risk stratification.

  2. Pulmonary Embolism Risk Stratification from CTPA and Medical Records: Vascular Graphs Are Not All You Need

    cs.CV 2026-06 unverdicted novelty 5.0

    In a private dataset of 353 patients, medical records and cardiac biomarkers outperform vascular biomarkers and GNNs on vascular graphs for PE risk stratification, suggesting vascular graphs hold no discriminative inf...

Reference graph

Works this paper leans on

1 extracted references · cited by 1 Pith paper

  1. [1]

    VSMOD: A Vessel Segmentation and MODelization plugin for 3D Slicer,

    [1] M. des Ligneris, G. Husak, A. Aillet, G. Doll ´e, C. Frindel, and O. Merveille, “VSMOD: A Vessel Segmentation and MODelization plugin for 3D Slicer,”Journal of Open Source Software, vol. 10, no. 114, p. 8563, Oct. 2025

    2025