PACS: Instrumentation, Medical, and Data Systems

• PACS are integrated systems spanning astrophysical instruments, medical imaging archives, and computational frameworks, characterized by precise calibration and robust analysis.
• Key methodologies include advanced calibration techniques, machine learning for source extraction, and hybrid algorithm pipelines tailored to diverse domain requirements.
• Applications range from far-infrared astronomical surveys and clinical diagnostic imaging to simulation studies and multi-omics cancer subtype classification, driving cross-domain innovations.

Picture Archiving and Communication Systems (PACS) serve as foundational infrastructures across multiple scientific and engineering domains, notably in astrophysics (as an instrument: Photodetector Array Camera and Spectrometer), in clinical informatics (as medical image archives: Picture Archiving and Communication System), and as acronymic frameworks in computational and data sciences. This article surveys the principal instantiations of PACS in far-infrared space instrumentation, astrophysical survey data products, medical imaging informatics, and selected computational models, with an emphasis on system design, calibration, and domain-specific applications.

1. Herschel/PACS: Instrument Architecture and Key Capabilities

The Photodetector Array Camera and Spectrometer (PACS) was one of three primary instruments on ESA's Herschel Space Observatory, optimized for far-infrared (FIR: 55–210 μm) photometry and integral-field spectroscopy (Poglitsch et al., 2010). The instrument comprised two back-ends: a dual-band photometer (using silicon bolometer arrays for 70, 100, and 160 μm imaging) and a spectrometer (using Ge:Ga photoconductor arrays for 5×5 spaxel, R ~ 1000–4000 spectroscopy). The photometer delivered simultaneous two-band mapping over ∼1.75′×3.5′, while the spectrometer provided spectral cubes over a 47″×47″ field-of-view.

The PACS system design included:

• Cold Lyot stop and long-wave blocking filter for stray-light suppression.
• Dichroic and image-slicer-based beam splitting for concurrent photometry/spectroscopy.
• ^3He sorption coolers maintaining bolometer operation at 0.3 K.
• Focal-plane chopper for differential sky/internal calibration.

In flight, PACS achieved point-source sensitivities of 5–10 mJy (5σ/1 h) in photometric mode, with beam FWHMs of ∼5.6″ (blue), ∼6.8″ (green), and ∼11.4″ (red). The integral-field spectrometer covered the 55–210 μm regime with instantaneous spectral coverage (Δv ≈ 1500 km s⁻¹) and velocity resolutions of 75–300 km s⁻¹ (Poglitsch et al., 2010, Sagawa et al., 2010).

2. Calibration Methodologies and Photometric Stability

PACS photometric calibration leveraged repeated observations of five fiducial IR-bright stars, realizing band-specific response factors R(λ) which converted raw detector output to flux densities. The absolute photometric scale is tied to stellar models with 5 % uncertainty (Müller et al., 2016). Calibration block analyses demonstrated an inherent bolometer response stability of σ≃0.5 % (RMS) and ∼8 % peak-to-peak over the full mission; after correction for evaporator temperature $T_\mathrm{Ev}$, post-recycling drift $t_\mathrm{rec}$, and FPU temperature $T_\mathrm{FPU}$ dependencies, residual RMS decreased to 0.12–0.18 % (peak-to-peak <2–5 %) (Moór et al., 2013). This enabled the use of a single, time-independent response file for all science data.

Non-linearity corrections (pre-flight characterization per pixel) maintained bolometric linearity over >4 orders of magnitude (few mJy to >1000 Jy), and cross-calibration with solar-system standards validated the absolute flux scale across a wide dynamic range (Müller et al., 2016).

3. Source Extraction, Catalogs, and Point Spread Function Characterization

The PACS Point Source Catalogue (HPPSC, first in 2017 (Marton et al., 2017), updated as HPPSC2 in 2024 (Marton et al., 2024)) provides the principal astronomical legacy resource. Source extraction deployed hybrid pipelines: classical algorithms (SUSSEXtractor, DAOStarFinder, Haar-feature finders), deep convolutional networks (ResidualHerschelNet), and random forest–based spurious source rejection. Photometric accuracy is achieved at the ∼1 % level, with completeness and purity dependent on background structure and source flux. In HPPSC2, deep learning and ML-based QA reduced spurious rates to ~1–2 % while completeness rose 5–50 % over HPPSC1, with robust cross-matching to external catalogs (e.g., Hi-GAL, PEP) (Marton et al., 2024).

Precise knowledge of the PACS point spread function is essential for accurate photometry and deconvolution. By combining Vesta and Mars observations, deep empirical PSFs (dynamic range ≳10⁶) were constructed for all bands and scan speeds. PSFs consist of a central Gaussian, diffraction lobes, and power-law wings with α∼3–3.3, yielding FWHMs of 5.8–12.3″ depending on band and scan velocity (Bocchio et al., 2016).

4. Spectroscopic and Photometric Science Applications

4.1 Protostellar and Circumstellar Environments

PACS FIR spectroscopy enabled the first full CO ladder coverage ($J_\mathrm{up}=14\text{--}46$) in protostellar populations, revealing shock-heated, sub-thermally excited gas as the dominant emission mechanism rather than PDRs, with universal rotational temperatures and L(CO)–L_bol correlations robust to $>2$ orders of magnitude in luminosity (Manoj et al., 2012). In AGB and OH/IR stars, PACS+SPIRE range spectroscopy yielded constraints on gas mass-loss, dust continuity, and molecular excitation (e.g., CO, H$_2$O lines), with rigorous LTE and non-LTE modeling pipelines quantifying mass-loss histories and dust–gas couplings (Lombaert et al., 2010, Nicolaes et al., 2018).

4.2 Extragalactic and High-Redshift Systems

Far-IR photometry of $z>4$ quasars established the presence of massive, cold dust reservoirs and FIR luminosities ($>10^{13}\,L_\odot$), co-eval with black-hole accretion, using standard conversions: $$L_{\nu,\rm rest} = 4\pi\,D_L^2\,\frac{S_{\nu,\rm obs}}{1+z} \quad M_{\rm dust} = \frac{S_{\nu,\rm obs}\,D_L^2}{(1+z)\,\kappa_\nu\,B_\nu(T_{\rm dust})}$$ (Leipski et al., 2010).

4.3 Disk Physics and Star Formation

PACS photometry (especially at 100 and 160 μm) provides crucial leverage in radiative transfer SED models of circumstellar disks, constraining dust mass, vertical structure, and flaring properties. However, partial optical thickness at 100 μm implies that measured masses set lower limits (Olofsson et al., 2013).

5. PACS in Medical Informatics: Picture Archiving and Communication System

In medical imaging, PACS refers to Picture Archiving and Communication Systems—enterprise solutions adhering to the DICOM standard that manage the acquisition, archival, retrieval, and distribution of clinical images. Contemporary PACS frameworks (e.g., Dicoogle (Lebre et al., 2019), Transversal PACS Browser API (Lameira et al., 2024)) abstract DICOM services (C-STORE, C-FIND, C-MOVE), offer plug-in extensibility (indexing, storage backends, RESTful services), and increasingly support cross-platform web-based viewers and advanced search. Key architectural features include plugin-managed indexing (e.g., Lucene), storage abstraction (local/filesystem/cloud), and query frameworks supporting custom DICOM tag searches and credentialed, role-based access.

Interoperability remains a challenge due to heterogenous vendor extensions, patient ID ambiguities, and varying DICOM tag sets. The Transversal PACS Browser API addresses multi-station querying, embedded JPEG previews, and advanced filtering (study/series/patient level) in a Qt/DCMTK stack, with performance validated via Likert-scale stakeholder surveys (Lameira et al., 2024). Open-source PACS solutions differ in plugin architecture, REST/DICOMWeb support, web visualization, and digital pathology integration; Dicoogle and Orthanc rank highest for extensibility and research applications (Lebre et al., 2019).

6. PACS as Computational and Data-Driven Frameworks

PACS is also used to denote specialized computational architectures and datasets:

• Polymer-Attenuated Coulombic Self-Assembly (PACS): An experimental and simulation framework for colloidal crystal formation using charged spheres with stabilizing polymer brushes. The effective pair potential combines short-range brush repulsion and tunable, screened Coulomb attraction:

$$V_\mathrm{total}(h) = V_\mathrm{brush}(h) + U_{DH}(h)$$

PACSim leverages OpenMM, supporting dynamic protocol updates and enhanced-sampling for nucleation studies (Höllmer et al., 13 May 2026).

• Prediction and Analysis of Cancer Subtypes (PACS): A Siamese multi-head attention model integrating multi-omics data (e.g., mRNA, methylation, miRNA, single-cell) for supervised cancer subtype classification. The architecture fuses encoder outputs via elementwise multiplication, enhancing non-linear feature interaction without increasing parameter count, and achieves state-of-the-art performance across simulated, single-cell, and TCGA datasets (Pan et al., 2023).
• PACS in Longitudinal CBCT Registration-Segmentation: Patient-specific Anatomic Context and Shape prior (PACS) combines recurrent unsupervised registration and one-shot segmentation of thoracic cone-beam CTs using CLSTM-based 3D U-Nets, outperforming baseline and multi-shot methods in tumor and organ-at-risk delineation (Jiang et al., 2022).
• PACS Dataset for Physical Audiovisual Commonsense: The PACS benchmark consists of ∼13,400 QA pairs over 1,526 videos, designed for evaluating audiovisual physical commonsense reasoning in AI. Baseline and state-of-the-art multimodal Transformer models underperform humans by >25 % absolute accuracy, underscoring the challenge of integrating audio and video information in multi-step physical reasoning (Yu et al., 2022).

7. Scientific and Community Impact

PACS, as instantiated in the Herschel mission, has produced foundational FIR survey catalogs, spectroscopic diagnostics, and calibration resources accurate to 1–5 %, forming the cross-wavelength calibration backbone for ALMA, SOFIA, future FIR observatories, and model refinement for planetary, disk, and extragalactic science (Marton et al., 2024, Müller et al., 2016).

In medical and computational contexts, open PACS frameworks and datasets drive innovations in healthcare interoperability, plug-in based clinical research, and multimodal data science. The proliferation of ML/AI PACS variants attests to evolving specialization and cross-domain synergy.

The diversity of systems and datasets denoted by PACS necessitates precise contextual usage; in all cases, technical rigor, standardized data interfaces, and robust calibration protocols remain central to effective deployment and scientific exploitation.