Background

• CT technology advances, such as AI-based reconstruction (mostly trained with patient data), present a wide variation of algorithms among manufacturers and little public information about their training datasets and how they operate [1].
• To guarantee an optimal use of such technologies with their potential dose reduction for patients, the use of anthropomorphic phantoms should be boosted in CT quality control and image quality evaluation, tracking and optimization [2].  Such phantoms, though some are available in the market, are not widely implemented in routine image quality optimization, due to their limited representation of a wide cohort of patients or disease variability.
• Lung cancer is the second most frequent type among men and women and the first one in cancer related projected deaths in the USA (2025), with similar trends worldwide [3]. Several lung cancer CT screening programs are active, including a wide range of CT systems and manufacturers and sometimes incorporates AI-based software for nodule detection and classification [4,5].  This represents a clear clinical indication which could be mimicked using anthropomorphic phantoms with lung nodules.
• 3D-printing has been used as an affordable manufacturing technology to create customized anthropomorphic phantoms in Radiology [6-9]
  

  Fig 1: Thorax PMMA phantom with two 3D-printed lung vessel inserts (Hernandez-Giron ECR 2024)
  .
• In this study, lung nodules were modelled following international guidelines and a wide range of materials and techniques were evaluated to produce them. The nodules were combined with commercial and 3D printed lung phantoms [10] and tested under CT imaging in terms of X-ray attenuation and morphometry.

Methods

Lung nodules' modelling

Three families of lung nodules’ and lung opacities models were created, based on:

• Patient CT-data segmentation from The Cancer Imaging Archive (TCIA-LIDC-IDRI, 5 nodules) in 3D-slicer by thresholding [11]
  

  Fig 2: Models of lung nodules extracted from CT patient data in The Cancer Imaging Archived (with their respective IDs), diagnosed as lung cancer. The largest nodule is shown on the right together with one axial CT image of it.
  . A Python module, pylidc [12] was adapted for querying the dataset and searching for subsets scans based on parameters of interest, which were high resolution CT and available patient diagnosis.
• Mathematical models (Meshmixer) (Fig. 3) based on simple shapes. For the latter, archetypes of varying complexity (sphere, ellipsoid, lobulated, spiculated, interlocking hollowed hemispheres (which allowed insertion of a core of a different material) were generated. These were rescaled following the British Thoracic Society Guidelines, which classifies lung nodules based on their volume (80, 125, 195, >300mm3), related to different diagnostic outcomes and patient management. Additionally, some of these models were generated with a hollowed interior to explore the potential of heterogeneous powder-based printing.
• Based on previous research [10] two lung vessel inserts were designed with selected closed regions to trap 3D-printed powder inside to mimic lung opacifications, which were also evaluated. 

3D-printing of lung nodules and material selection

In a pilot study (printing solid and hollowed cubes), it was determined that for selective laser sintering (SLS) printers and HP Multi Jet Fusion printers offered a wide variation of attenuation between their raw powder and its solid printed form (around 600 HU difference between both states). This was used to introduce lung opacities in  3D printed lung vessel phantoms ((150x100x60mm) based on a mathematical model of vessels distribution informed by patient data [7,10,13]) creating vessel traps in selected regions. Two lung vessel inserts were printed this way, selecting HP MJF and TPU (density=1.10g/cm³) which is a flexible and durable material, facilitating nodule placement. This material was previously noted as a good surrogate of blood vessels without contrast (~60 HU at 120kV in CT). 

Twenty nodules were printed for each geometry for a range of materials (HP-MJF:PA11-PA12-PP; SLS:PA11-PA12, PRUSA(orange/gray-resin)) to evaluate accuracy and reproducibility of the printing technologies. Two external 3D printed company (materialise [14], Sculpteo [15]) was used for all the prints (except resin printing, done in-house) and objects were printed in two batches to check reproducibility.

Evaluation of morphometry of 3D-printed lung nodules: CT imaging

The nodules were organized in polystyrene boxes, labelled, and scanned in a clinical CT system (standard-thorax-protocol, Siemens Somatom Edge Plus-CT). The images were inspected visually to perform an initial quality control of the prints and discard nodules with evident artifacts (like undesired airgaps not present in design).

Nodules were segmented in the images (3D-slicer), and attenuation measurements performed (average CT value inside the segmented volume). The segmented volumes were converted into STLs and registered to the original models to perform a morphometry analysis (MeshInspector: volume, average absolute distance, coloured distance maps).

Combination of lung nodules and anthropomorphic commercial and 3D-printed lung phantoms: CT imaging

A selection of the successfully printed nodules was attached at different locations in a:

• Custom PMMA thorax phantom (300x200x60mm) with the two 3D-printed lung inserts [10] were placed in a thorax PMMA shaped holder (300x200x60mm) which is equivalent to a teen or thin adult patient in terms of attenuation (which will be called 3D-printed custom phantom now on).  
• Commercial anthropomorphic phantom (Kyoto Kagaku paediatric chest phantom [16]) representing a 5-year old and its lung vessel distribution (Fig YY). The nodules were attached with double sided tape in both cases, though in this case, their weight make their placement challenging.

Both phantoms were scanned in a clinical Philips Revolution CT with the adult and paediatric standard thorax protocols, respectively. The attenuation of the lung nodules (average CT value in segmented volumes) and their appearance under usual reconstruction thickness (0.625mm, 2.5mm, lung and mediastinum).