PaperlayerView exact revision on arXiv

Source-linked AI summary

Feasibility of a Patient-Specific Digital Twin for Ventriculoperitoneal Shunt Trajectory Planning

Kritsanavis Chongsrid, Vich Yindeedej, Supaporn Konmun, Xulei Yang, Si Yong Yeo

arXiv:2610.20074v1cs.CVq-bio.QM

TL;DR

Catheter obstruction is the commonest cause of VP shunt failure, and placement is ordinarily determined freehand from skin landmarks. This study uses a deterministic CT-based digital twin to plan a patient-specific trajectory; in 11 adults, Grade 1 placement was achieved in 10, compared with 22 of 213 historically.

  • Problem

    Catheter obstruction is the commonest cause of shunt failure, and standard landmark-based entry and hand-held direction do not tailor the trajectory to the patient’s anatomy.

    Source: Introduction · p. 1

  • Method

    A deterministic digital twin searches preoperative CT for a safe perpendicular trajectory maximizing intraventricular catheter length and returns ruler measurements and depths for freehand execution.

    Source: Introduction · p. 2

  • Results

    10 of 11 patients achieved Grade 1 placement, against 22 of 213 (10.3%) in the historical freehand Keen’s point series.

    Source: Discussion · p. 10

  • Takeaways & Limitations

    The plan was executed freehand using two ruler measurements and two depths, with the gain attributed to patient-specific entry and depth rather than a change in insertion technique.

    Source: Discussion · p. 10

    This single-centre series included 11 patients, and follow-up was limited to immediate postoperative CT, so shunt survival, occlusion, and revision were not assessed.

    Source: Discussion · pp. 11–12

Abstract

from arXiv · show

Catheter obstruction is the commonest cause of shunt failure; freehand insertion at Keen's point leaves the tip free in cerebrospinal fluid in only one case in ten. We built a digital twin that reconstructs skull, scalp, ear and ventricles from preoperative CT by deterministic geometry, searches every entry around Keen's point for the perpendicular trajectory with the longest intraventricular length under safety constraints, and returns the plan as two ruler measurements from the ear and two depths. Executed freehand in 11 consecutive adults (IDEAL Stage 2a), it achieved Hayhurst Grade 1 placement in 10 of 11 against 22 of 213 historically (p = 9 x 10^-9), median entry and angular deviations 2.5 mm and 5.9 degrees, 1.4 cm more intraventricular catheter than Keen's point, and no intraoperative complication. The technique was unchanged; the gain came from choosing entry and depth per patient. Whether it lowers revision is the next stage's hypothesis.

Introduction

Catheter position matters because obstruction is the commonest cause of shunt failure, yet standard landmark-based freehand planning does not tailor entry, direction, and depth to each patient. The study introduces a CT-based digital twin and evaluates its freehand execution in 11 adults before testing whether planning reduces shunt failure.

Introduction

  • Catheter obstruction is the commonest cause of shunt failure, and the catheter’s position determines that risk.The perforated inlet segment must lie free in cerebrospinal fluid to avoid occlusion against ventricular structures.Source: Introduction · p. 1
  • Standard entry points are fixed offsets from external landmarks rather than choices adapted to the individual ventricle or skull, although entry largely determines catheter endpoint and parenchymal traversal.The catheter is then passed perpendicular to the skull.Source: Introduction · p. 1
  • Navigation and related execution aids can improve placement, but they require equipment, time, and training that many centres lack, while no step evaluates candidate trajectories against the patient’s ventricle.Metaanalyses show little or no effect on revision.Source: Introduction · p. 2
  • The digital twin searches around Keen’s point for a perpendicular entry maximizing continuous intraventricular catheter length under safety constraints, then gives ruler measurements and depths for team review.It uses deterministic geometry from routine preoperative CT and requires no intraoperative hardware.Source: Introduction · p. 2
  • The study evaluates a patient-specific CT plan executed freehand in 11 consecutive adults, with Grade 1 placement on postoperative CT compared against the institution’s historical Keen’s point series.The work is an IDEAL Stage 2a development study before multicentre validation and a comparative trial of shunt failure.Source: Introduction · p. 2

Results

In 11 consecutive adults, the patient-specific plan met its geometric safety constraints and was executed with high placement accuracy and no intraoperative complications. Against historical freehand Keen’s point insertions, it had a higher Grade 1 placement rate; geometric comparisons also found more intraventricular catheter length than landmark trajectories.

Patient flow and characteristics

  • The cohort comprised 11 consecutive adults; 12 of 17 scheduled patients had plans executed or considered, while clinical decisions and eligibility excluded others.Two patients were outside plan conditions, three had another course chosen clinically, and one planned patient had the insertion side changed because of skin infection.Source: Patient flow and characteristics · pp. 2–3

Plan generation

  • All 11 generated plans met the predefined constraints: D1 ≤4.0 cm, D2 ≥2.5 cm, and D ≥6.0 cm.Observed ranges were 2.1–3.7 cm for D1, 3.2–5.3 cm for D2, and 6.4–7.6 cm for D.Source: Plan generation · p. 3

Primary outcome: catheter placement accuracy

Execution fidelity

  • Nine of 11 cases had entry-point deviation <5 mm, and the median composite angular deviation was 5.9° (IQR 4.4–7.6°).The composite median was below the geometric (10.5°) and execution (9.5°) angular deviation medians.Source: Table 1 · p. 4, Execution fidelity · p. 5

Safety outcomes

  • No intraoperative complications occurred, and no executed catheter contacted the thalamus, caudate, or other structures outside its intended path.Postoperative imaging showed no haemorrhage along the track, choroid plexus contact, or ependymal penetration; the Case 2 tip contacted the septum pellucidum.Source: Safety outcomes · p. 6

Geometric counterfactual analysis

  • The computed trajectory provided a median 1.4 cm more intraventricular catheter than Keen’s point at 6 cm, with no algorithm trajectory extending beyond the ventricle.The Keen’s point trajectory fell below the 2.5 cm intraventricular floor in 3/11 cases at 6 cm; the algorithm’s advantage persisted at equal planned depth.Source: Geometric counterfactual analysis · p. 8, Table 2 · p. 8
  • The modified Keen’s point trajectory had a median intraventricular length of 1.4 cm, versus 4.1 cm for the algorithm; its tip lay outside the ipsilateral ventricle in 10/11 cases.The algorithm-versus-modified-Keen’s median difference was 2.3 cm (11/11, Wilcoxon signed-rank p=0.0010).Source: Geometric counterfactual analysis · p. 8, Figure 5 · p. 9

Comparison to historical controls

  • 90.9% (10/11) achieved Grade 1 placement versus 10.3% (22/213) in the historical freehand Keen’s point series (p = 9 × 10^-9).Good tip location was also more frequent: 10/11 versus 41/213 historical cases (p = 1.6 × 10^-6).Source: Comparison to historical controls · p. 9, Figure 6 · p. 10

Discussion

The patient-specific plan improved placement over historical freehand Keen’s point insertion while leaving the insertion technique unchanged. The study establishes technical feasibility, not a reduction in revisions; broader testing and follow-up are planned.

Discussion

  • 10 of 11 patients had Grade 1 placement, versus 22 of 213 (10.3%) in the institution’s historical freehand Keen’s point series.Median entry and angular deviations were 2.5 mm and 5.9°.Source: Discussion · p. 10
  • The insertion technique stayed the same; the plan changed the entry and depths, while a 2.5 mm entry shift on the curved skull was compensated by hand.The composite angular deviation was 5.9°, smaller than either component deviation (9.5° and 10.5°).Source: Discussion · p. 10
  • 1.4 cm more intraventricular catheter was achieved than with the Keen’s point trajectory on the same 11 ventricles, meeting the 2.5 cm floor in every case.Keen’s point failed to meet the floor in three cases and left the ventricle in four.Source: Discussion · p. 10
  • The planned placement target is tied to survival in adults, but whether this placement gain improves survival remains a hypothesis for the next study stage.The study assessed immediate placement, not shunt survival or revision.Source: Discussion · pp. 11–12
  • The comparison is limited by 11 patients at one centre, a wide 62 to 98% confidence interval, and non-randomized historical controls graded by different readers.Postoperative imaging was obtained at different intervals.Source: Discussion · p. 11
  • The study’s scope is adults with Evans index ≤0.45, eligible ventricular geometry and corridor, an unfolded pinna on CT, and thin-slice CT from one scanner.Follow-up was limited to the immediate postoperative scan.Source: Discussion · p. 12
  • An IDEAL Stage 2b multicentre study is planned to test Grade 1 placement across centres and scanners, measure surgeon-specific execution fidelity, and follow placement-to-survival outcomes for at least one year.A planned septum and wall clearance constraint addresses the single Grade 2 placement, where full-depth rigid insertion brought a tip near the septum into contact.Source: Discussion · p. 12

Methods

The system reconstructs a patient-specific twin from routine preoperative CT, searches a posterior parietal entry window, and reports ruler measurements and catheter depths. The plan is executed freehand after the team reviews the alignment and segmentation.

Overview

  • The planning system automatically constructs a static patient-specific twin from preoperative CT and returns entry measurements A and B plus depths D1 and D2; the team reviews alignment and segmentation before the search.The study also analyzes preoperative and postoperative CT in 11 consecutive patients whose plans were executed freehand.Source: Overview · p. 12

Setting and eligibility

  • The plan was used for posterior parietal insertion when suitable preoperative CT was available, the pinna was not folded against the scalp, and Evans index was 0.45 or less.The route was the institution’s default approach.Source: Setting and eligibility · p. 12

Digital twin construction

  • Thresholding and rule-based selection segment skull, scalp, ear, and ventricles; surfaces are placed in a common head frame defined by MNI alignment and the Frankfurt plane.The ear segmentation assumes the pinna stands off the head, and the mid-sagittal plane divides the ventricles into ipsilateral and contralateral sides.Source: Digital twin construction · p. 13

Trajectory generation and ranking

  • The inward normal at each inner-table vertex defines the candidate trajectory, while A and B index entries by height above the helix and posterior scalp arc length.Candidates span A 2.0–4.5 cm and B at least 2.0 cm; the window brackets classical Keen’s point.Source: Digital twin construction · p. 13, Trajectory generation and ranking · pp. 13–14, Figure 7 · p. 14
  • Candidates are ray-tested against safety constraints, then passing paths are rehearsed as a catheter-sized cylinder within the ipsilateral ventricle and ranked by effective intraventricular length D2.Constraints include atrial entry, parenchymal traversal no greater than 4.0 cm, at least 2.5 cm intraventricular length, insertion depth of at least 6.0 cm, and no posterior direction component.Source: Trajectory generation and ranking · pp. 14–15, Figure 8 · p. 16
  • The plan reports A and B from the pinna reference and D1 and D2, whose sum is insertion depth from the inner table to the tip.The application runs on a hospital computer without requiring a Python installation.Source: Trajectory generation and ranking · p. 15

Algorithm 1. Trajectory search

The planning algorithm searches patient-specific skull entry points for a perpendicular trajectory meeting ventricular-length constraints, then gives the team ruler measurements and insertion depths. The study assesses execution, postoperative placement, landmark comparators, and a historical control.

Algorithm 1. Trajectory search

  • The search filters inner-table vertices by a ray’s ventricular entry, parenchymal traversal, and length constraints, then verifies candidates with a catheter-sized cylinder and selects the longest feasible intraventricular segment.It searches first within B ≤ 4.5 cm, then beyond that band, and returns “no feasible trajectory” if neither yields a candidate.Source: Algorithm 1. Trajectory search · p. 17

Surgical execution

  • The team measures A and B from the ear and inserts the catheter perpendicular to the skull to planned depth D, without navigation, ultrasound, intraoperative imaging, or an insertion guide.The rigid cannula creates a straight track, and the catheter follows it; planned-versus-executed comparisons assume a straight line from entry to tip.Source: Surgical execution · p. 17

Postoperative imaging and catheter reconstruction

  • Postoperative CT is rigidly registered to preoperative CT; the catheter centreline is reconstructed from its barium signal, with a manually placed entry-to-tip line as the only manual step.A catheter-diameter tube is swept along the centreline; the executed tube has no safety margin, unlike the plan.Source: Postoperative imaging and catheter… · p. 18

Outcome measures

  • Placement grade is assessed on postoperative CT, while secondary measures cover trajectory fidelity, distances to 47 structures, temporalis contact, and specified safety outcomes.Hayhurst grades distinguish a free-floating tip (Grade 1), wall or choroid contact (Grade 2), and partial parenchymal placement or failed cannulation (Grade 3); safety outcomes include intraoperative complications, track haemorrhage, choroid contact, and ependymal penetration.Source: Outcome measures · p. 18

Keen’s point comparators

  • Classical Keen’s point is constructed from two 3 cm scalp measurements and tested at 6 cm and 8 cm depths; the modified point is 4 cm superior and posterior to the ear canal and tested at 5 cm and 6 cm.A sensitivity analysis also compares classical Keen’s point at the algorithm’s patient-specific planned depth.Source: Keen’s point comparators · pp. 18–19

Historical controls

  • The historical comparator comprises 213 consecutive freehand Keen’s point insertions at the same institution, graded from routine postoperative imaging; thicker slices in that series may make Grade 1 comparisons favor the historical cohort.No patient-level data from the historical series were used.Source: Historical controls · p. 19

Statistics and reproducibility

  • With 11 versus 213 patients, the study uses Fisher’s exact tests for historical comparisons and Wilcoxon signed-rank tests for paired comparisons, without multiple-comparison adjustment.As an IDEAL Stage 2a development report, the sample was the consecutive series rather than a powered sample; p values are descriptive statistics.Source: Statistics and reproducibility · p. 19

Use of large language models

  • Large language models assisted with drafting and editing, but authors verified statements, numbers, and references; no generative AI produced data or images.Source: Use of large language models · p. 19

Data availability

Per-case plan and counterfactual values are reported in tables, while imaging access is restricted.

Data availability

  • Per-case plan and counterfactual values are provided in Table 2 and Supplementary Tables 2 and 3.Source: Data availability · p. 19
  • Imaging data are not public but may be requested from the corresponding authors, subject to ethics approval and a data use agreement.Source: Data availability · p. 19

Code availability

The source code is not publicly available, but the method is specified in enough detail for reproduction from the description.

Code availability

  • The planning system’s source code is not publicly available because it is subject to a joint technology disclosure.Source: Code availability · p. 19
  • The algorithm and its constraints and parameters are specified in Methods and Supplementary Note 1 for reproduction from the description.Source: Code availability · p. 20
  • Non-commercial verification access requests will be considered by the corresponding authors, subject to agreement among the three institutions.Source: Code availability · p. 20

Competing interests

The authors report a technology disclosure naming four inventors and no other competing interests beyond that disclosure.

Competing interests

  • Four authors are named inventors on a joint technology disclosure covering the planning system; no patent application had been filed at submission.Source: Competing interests · p. 20
  • One author declares no competing interests, and the others report none beyond the technology disclosure.Source: Competing interests · p. 20

Supplementary Note 1. Formal specification of the planning system

The formal specification reconstructs anatomical masks, defines the atrium and evaluates candidate trajectories through ray and volumetric checks. It selects a feasible plan using safety constraints and a staged search around Keen’s point.

Supplementary Note 1. Formal specification of the planning system

  • The lateral-ventricle mask is formed from intracranial voxels in the cerebrospinal-fluid range, then refined by erosion, component selection, dilation and closing.Erosion separates the lateral ventricles from the third ventricle at the interventricular foramen; dilation and closing restore the margin and fill holes.Source: Supplementary Note 1. Formal… · p. 23

Supplementary Figure 1. The atrium label

  • The atrium is operationally defined as the ventricular region between two coronal planes tangent to the largest inscribed circle in the reference cross-section.The supplementary figure labels the body anterior to the planes, the atrium between them and the occipital horn posterior to them.Source: Supplementary Note 1. Formal… · p. 24, Figure 1 · p. 24
  • For each candidate vertex, the ray construction uses its inward unit normal and the ipsilateral lateral ventricle.The supplied formal specification introduces the ray in these terms but does not provide further readable construction details.Source: Supplementary Figure 1. The atrium label · p. 24
  • The catheter’s volumetric check uses a stack of discs whose radius is rounded up to whole voxels, making the collision check conservative.Effective intraventricular length begins at the depth where the modelled catheter first lies wholly within the ventricle.Source: Supplementary Figure 1. The atrium label · p. 24
  • The planning sequence includes a voxel-scale bound for early stopping, with the ray and cylinder read on the same axis samples.The specification states that the bound holds for every candidate.Source: Supplementary Figure 1. The atrium label · p. 25

Supplementary Figure 2. Why a plan comes from the second band

  • The candidate window is searched first between the ear and B = 4.5 cm, then beyond that line only if the first band yields no feasible candidate.Candidates are classified by the first constraint they fail, including parenchymal traversal, atrial entry, intraventricular length and volumetric checks.Source: Figure 2 · p. 25
  • Only the second band yielded a plan: none of 4,654 first-band candidates passed every constraint, and the selected entry was at B = 5.3 cm.The full candidate window contained 16,429 candidates; four first-band candidates passed the first four constraints but lost intraventricular length at the volumetric check.Source: Figure 2 · p. 25, Supplementary Figure 2. Why a plan… · p. 26

Supplementary Note 2. Implementation

The system is packaged as a standalone Windows application with a graphical interface for inspecting the plan. Supplementary materials show planned and executed trajectory directions, coordinate measurements, and comparisons with a modified Keen’s point trajectory.

Supplementary Note 2. Implementation

  • The system uses Python scientific-imaging libraries and a PyQt5 interface, and is compiled into a standalone Windows application.An optional GPU path runs cylinder construction; computation times were measured on a hospital computer with a laptop-class processor.Source: Supplementary Note 2. Implementation · p. 26

Supplementary Figure 3. User interface of the planning system

  • The interface pairs axial, sagittal and coronal views with trajectory-aligned views and a 3D render of the ventricle and planned trajectory.The render labels A, B, D1, D2 and D; the ray and catheter segments use corresponding colors across the views.Source: Figure 3 · p. 26

Supplementary Figure 4. Azimuth and elevation of a trajectory

  • Executed directions differed from planned directions by a median +0.1° in azimuth and −2.7° in elevation, with no systematic sideways bias.Planned and executed directions pointed forward and inward, 30° to 57° from straight medial, and lay within 12° of the Frankfurt plane.Source: Table 1 · p. 26, Supplementary Figure 3. User interface… · p. 27, Figure 4 · p. 27
  • The deviation schematic separates execution, geometric and composite angles, alongside entry, tip and catheter-length deviations.The angles share a vertex after translating the planned direction to the executed entry; the schematic notes that cohort medians are in Table 1.Source: Figure 5 · p. 28
  • At the median, executed measurements differed from planned values by 0.8 mm in A, less than 0.1 mm in B, 0.0 mm in D1, 0.1 mm in D2 and 0.4 mm in D.A and B locate entry; D is entry-to-tip distance, D1 is the pre-ventricular part and D2 is the intraventricular part.Source: Supplementary Figure 4. Azimuth and… · pp. 28–29, Table 2 · p. 28
  • The modified Keen’s trajectory left through a ventricular wall into white matter in six cases, the hippocampus in three and the thalamus in one; none crossed the septum pellucidum.At 6 cm, the tip lay within the lateral ventricle in one case; the additional centimetre from 5 to 6 cm added only overshoot in every case.Source: Supplementary Figure 4. Azimuth and… · p. 29, Table 3 · p. 29

Related papers

Loading 2610.20074v1…