Ep 37 — FPA Focal Plane Array Imaging: High-Throughput Chemical Imaging

Series: Infrared Spectroscopy Encyclopedia: From Principles to Practice
Chapter: Part IV · Advanced — Frontier Technologies (Episode 2)
Audience: Microplastic research and environmental monitoring personnel, biomedical imaging researchers, polymer blend analysis engineers, chemometrics/spectral data processing personnel
Prerequisites: Ep 14 (ATR principles), Ep 26 (Microplastic detection), Ep 33 (Tissue infrared diagnosis), Ep 36 (Micro-FTIR basics)
Reading time: Approximately 50 minutes

Introduction: One filter, 470 GB of data, 110,000 microplastics

In 2022, Wang's team from Qingdao University published a study on microplastics from Chinese takeaway containers in Molecules [1]:

"Using FPA-based μ-FTIR whole-filter analysis, we identified 110,000+ microplastic particles on a single 47 mm membrane filter from a single takeaway container, with sizes ranging from 11 μm to 5 mm. The 6-hour measurement generated ~470 GB of raw spectral data, automatically processed by siMPle software to classify 9 polymer types."
—— Zhou X, Wang J, Ren J. Molecules 2022 [1]

This case demonstrates the remarkable capability of FPA (Focal Plane Array) imaging: it condenses what used to require weeks of manual point-by-point identification into a few hours of automated imaging, pushing chemical imaging from a "scientific luxury" to a "high-throughput routine tool." This episode will systematically explain the FPA detector principle, data cube structure, speed comparison with single-point MCT, and through two typical cases—microplastic whole-filter analysis and tissue section imaging—show how FPA reshapes the workflow of infrared chemical imaging, and discuss the big data and chemometrics challenges it brings.

1. FPA Detector: From "Thousand Measurements" to "One Exposure"

1.1 Bottleneck of Single-Point MCT

Recall Ep 36: Traditional μ-FTIR uses a single-point MCT (Mercury Cadmium Telluride) detector. To construct a chemical image, it must perform point-by-point scanning—the XY stage moves one step, one spectrum is measured, then moves again, another spectrum, and so on [2][3].

For a 1 mm × 1 mm area with 5 μm step size:

    Points = (1000/5) × (1000/5) = 200 × 200 = 40,000 points
    Scanning time per point = 30 s (128 scans, 4 cm⁻¹ resolution)
    Total time = 40,000 × 30 s = 1,200,000 s ≈ 333 hours ≈ 14 days

14 days to measure 1 square millimeter—that is the "throughput bottleneck" of single-point MCT in chemical imaging. At such throughput, imaging a whole microplastic filter (47 mm diameter) is practically impossible [2][3].

1.2 FPA: "Parallel Measurement" with a 2D Array

The core idea of FPA: Integrate thousands of detector pixels on the focal plane and acquire all pixel spectra simultaneously in one exposure, effectively executing thousands of single-point measurements in parallel [3][4][5].

   Traditional Single-Point MCT:       FPA Focal Plane Array:

   Sample                             Sample
     ↓                                  ↓
   Aperture (1 point)                 Objective (imaging)
     ↓                                  ↓
   MCT detector                       Focal plane (4096 points)
     ↓                                  ↓
   1 spectrum / exposure              4096 spectra / exposure
     ↓                                  ↓
   XY scan 40,000 times              XY scan 10 times
     ↓                                  ↓
   333 hours                         ~30 minutes

Physical structure of FPA (taking 64×64 array as an example) [3][4]:

   ┌─────────────────────────────────────┐
   │  FPA Array (64×64 = 4096 pixels)    │
   │                                     │
   │  ● ● ● ● ● ● ● ● ● ● ● ● … ● ● ●  │  ← each pixel is an
   │  ● ● ● ● ● ● ● ● ● ● ● ● … ● ● ●  │     independent MCT detector
   │  ● ● ● ● ● ● ● ● ● ● ● ● … ● ● ●  │     pixel pitch 40–50 μm
   │  ● ● ● ● ● ● ● ● ● ● ● ● … ● ● ●  │     (equivalent on sample
   │  ● ● ● ● ● ● ● ● ● ● ● ● … ● ● ●  │     5.5 μm @ 15× objective)
   │  …                                  │
   │  ● ● ● ● ● ● ● ● ● ● ● ● … ● ● ●  │
   └─────────────────────────────────────┘
                ↓
   IR light is imaged onto the array via objective
   Each pixel corresponds to a small area on the sample

1.3 Key Parameters of FPA

Table 1: Key parameters of FPA detector (Data sources: Bruker Hyperion Series Manual [4]; Agilent 620 FTIR Manual [5])

1.4 Spectral Acquisition Principle of FPA

The key difference between FPA and single-point MCT lies in the signal readout method [3][4]:

1. Interferometer scanning: The FTIR interferometer's moving mirror moves, generating an interferogram—identical to single-point MCT;
2. Array readout: At each sampling point of the interferogram (i.e., each position of the moving mirror), all 4096 pixels of the FPA array read out one frame simultaneously—each pixel records the interferogram signal at that position at that moment;
3. Data cube construction: After the moving mirror completes scanning, each pixel obtains a complete interferogram, which is Fourier transformed to yield one spectrum, ultimately building a three-dimensional data cube (detailed in Section 3).

Key performance feature: The "parallelism" of FPA comes from reading out 4096 pixels in one exposure, not from increasing the interferometer scan speed. Thus, FPA uses the same interferometer as single-point MCT, but the imaging speed is increased by a factor of 4096 (theoretical), with actual improvements of about 40–100 times due to readout overhead and SNR adjustments [3][4].

II. Chemical Imaging: Every Pixel is a Complete Spectrum

2.1 Concept of Chemical Imaging

Conventional optical imaging (including visible microscopy, SEM/EDS) provides morphological information—each pixel in a two-dimensional image records a physical quantity (brightness, color, elemental abundance). In contrast, FPA infrared chemical imaging is characterized by [3][6]:

Each pixel carries a complete infrared spectrum—this is the true meaning of "chemical imaging."

This means that from a single FPA dataset, users can:

• Extract the spectrum of any pixel for library search to identify the material at that location;
• Select any characteristic peak (e.g., 1715 cm⁻¹ ester C=O) and plot the intensity distribution of that peak to obtain an "ester distribution map";
• Perform chemometric classification (PCA, PLS-DA) on each pixel to obtain a "chemical class image";
• Extract spectra from different locations for comparison to study the chemical heterogeneity of the sample.

2.2 Presentation Modes of Chemical Imaging

FPA chemical imaging is typically presented in three ways [3][6]:

1. Chemical Map (Single Peak):
Select a characteristic wavenumber (e.g., 1715 cm⁻¹), map the absorbance at that wavenumber for each pixel to a color, and obtain a two-dimensional distribution map of that compound. Simple and intuitive, but ignores peak shape and overlap information.

   ┌──────────────────────────────────────┐
   │  1715 cm⁻¹ Chemical Map (Ester C=O Distribution)       │
   │                                       │
   │     ░░░░░░░░░░░░░░░░░░░░░░░░░░░     │  ← Red (strong): high concentration
   │     ░░░▓▓▓▓▓░░░░░▓▓▓░░░░░░░░░░     │
   │     ░░▓▓▓▓▓▓▓▓░░░▓▓▓▓▓░░░░░░░░     │  ← Yellow (medium): medium concentration
   │     ░░▓▓▓▓▓▓▓▓░░░▓▓▓▓▓▓░░░░░░░     │
   │     ░░░▓▓▓▓▓░░░░░░▓▓▓░░░░░░░░░░     │  ← Blue (weak): low concentration
   │     ░░░░░░░░░░░░░░░░░░░░░░░░░░░     │
   └──────────────────────────────────────┘

2. Class Image:
Perform library search or chemometric classification on each pixel, assigning different colors to different materials. For example, in microplastic analysis, PE is colored red, PP blue, PS green, etc.

3. PCA Score Map:
Perform PCA on the data cube, map the scores of each pixel on principal components to colors, revealing the chemical clustering structure of the sample without prior knowledge of the components.

2.3 Application Levels of Chemical Imaging

Table 2: Application levels of FPA chemical imaging (data source: Bhargava Spectroscopy 2013 [6])

III. Data Cube: Structure of Three-Dimensional Data

3.1 Definition of Data Cube

FPA imaging produces a three-dimensional data cube (hypercube) [3][7]:

              Wavenumber (λ) →
              ↓ 4000 3000 2000 1000 900 cm⁻¹
           ┌───────────────────────────┐
         y │  ●  ●  ●  ●  ●  ●  ●  ●  │  ← Each pixel position
         ↓ │  ●  ●  ●  ●  ●  ●  ●  ●  │    (x,y) stores a
           │  ●  ●  ●  ●  ●  ●  ●  ●  │    complete infrared spectrum
           │  ●  ●  ●  ●  ●  ●  ●  ●  │
           │  ●  ●  ●  ●  ●  ●  ●  ●  │
           └───────────────────────────┘
              x →

   Dimension 1: x (spatial, pixels)
   Dimension 2: y (spatial, pixels)
   Dimension 3: λ (wavenumber, spectral channels)

Mathematical expression: I(x, y, λ)—absorbance at position (x, y) and wavenumber λ.

3.2 Data Size Estimation

The size of a data cube depends on array size, spectral range, and resolution [3][7]:

Table 3: FPA data size estimation (data source: Primpke Appl Spectrosc 2020 [7])

Key observation: Single frame size grows linearly, but the area covered per frame also grows linearly, so for the same total area, the total data volume is roughly constant (~470 GB / 47 mm filter) [7].

3.3 Data Storage and Formats

Common data formats [3][7][8]:

• Instrument-specific formats: Bruker *.dsc / *.dmd (data cube + metadata); Agilent *.dmt / *.seq; PerkinElmer *.fsm;
• General formats: HDF5 (recommended by HyperSpy), ENVI, JCAMP-DX (single spectrum only), Matlab *.mat;
• Open tool support: HyperSpy, SpectroChemPy, OpenSpecy can all read major formats (see Ep 54) [8].

3.4 "Slicing" Operations on the Data Cube

The data cube can be "sliced" along three dimensions to obtain information from different perspectives [3][6]:

   Slice method 1: Spatial slice (single spectrum)
   → Slice along λ axis at (x₀, y₀) to obtain a spectrum

   Slice method 2: Spectral slice (single-peak chemical map)
   → Slice along the (x, y) plane at λ₀ to obtain a 2D chemical map

   Slice method 3: Wavenumber range integration (integrated chemical map)
   → Integrate each pixel over the range λ₁ to λ₂ to obtain a 2D chemical map

Practical experience: Integrated chemical maps (e.g., integration over 1700–1750 cm⁻¹) have higher SNR than single-peak chemical maps (only 1715 cm⁻¹) and are commonly used [6].

IV. FPA vs. Single-Point MCT: Comprehensive Comparison of Speed and Performance

4.1 Speed Comparison

Using "scanning a 1 mm × 1 mm area with 5 μm step size" as a benchmark [3][7]:

| 32×32 FPA | 1,024 | ~30 s per frame (full array parallel) | ~20 min | 1000× |
| 64×64 FPA | 4,096 | ~30 s per frame (full array parallel) | ~5 min | 4000× |
| 128×128 FPA | 16,384 | ~30 s per frame (full array parallel) | ~1.5 min | 13000× |

Table 4: FPA vs single-point MCT speed comparison (baseline: 1 mm², 5 μm step, 4 cm⁻¹, 128 scans)

Note: Actual speedup is affected by readout overhead, SNR adjustment, tiling time, etc.; theoretical values are typically reduced by 50–90%. Measured speedup of 64×64 FPA versus single-point MCT by Primpke et al. is 40–100× [7].

4.2 SNR Comparison

Table 5: SNR comparison of FPA vs single-point MCT (data source: Nasse Nature Methods 2011 [3])

Key tradeoff: FPA trades SNR for speed. For applications requiring extremely high SNR (e.g., secondary structure analysis of single-cell proteins), single-point MCT or synchrotron + single-point MCT remains the top choice; for throughput-driven applications (e.g., full-filter screening of microplastics), FPA is the only way to go [3][7].

4.3 Spatial Resolution Comparison

Both FPA and single-point MCT are limited by diffraction, so theoretically the same spatial resolution:

• Transmission mode: ~10 μm @ 1000 cm⁻¹ (diffraction limit)
• ATR mode (Ge crystal): ~3 μm @ 1000 cm⁻¹

However, the actual resolution of FPA is often limited by pixel sampling (Nyquist) — pixel size should be no larger than about λ/2. With a 15× objective, the pixel size is ~5.5 μm, which for λ=10 μm (1000 cm⁻¹) is at the border of near/under-sampling, often acceptable in practice; lower wavenumbers (longer wavelengths) are more prone to under-sampling, requiring a change in magnification or accepting resolution loss [3][7].

4.4 Selection Decision Tree

   Does your application need chemical imaging?
            │
            ├─ Yes → How large is the area?
            │       │
            │       ├─ < 100×100 μm  → Single-point MCT (small area, high SNR)
            │       │
            │       ├─ 100 μm–1 mm   → Single-point MCT or 32×32 FPA
            │       │
            │       └─ > 1 mm         → FPA (high throughput required)
            │
            └─ No (only point measurement) → Single-point MCT

5. Application Case 1: Automatic Analysis of Microplastics on Full Filter

5.1 FPA Workflow for Microplastic Detection

Microplastics (MPs) are the most classic application of FPA chemical imaging. Complete workflow (using Agilent 620 + 620-IRT + 64×64 FPA as an example) [1][7][9]:

   ┌─────────────────────────────────────────────┐
   │ Step 1: Sampling                            │
   │   Water/food/sediment → vacuum filter onto GF/F │
   │   (glass fiber, 47 mm diameter, 0.7 μm pores)  │
   ├─────────────────────────────────────────────┤
   │ Step 2: Digestion                           │
   │   Fenton reagent (H₂O₂ + Fe²⁺) to remove organic │
   │   matter, avoid biological debris interference│
   ├─────────────────────────────────────────────┤
   │ Step 3: Drying                              │
   │   60°C oven for 2 h to avoid water peak interference│
   ├─────────────────────────────────────────────┤
   │ Step 4: FPA Imaging                         │
   │   · Transmission mode (filter IR transparent)│
   │   · 15× objective, pixel 5.5 μm             │
   │   · 64×64 FPA, single frame 350×350 μm      │
   │   · Automatic tiling: ~19,000 frames cover 47 mm filter│
   │   · 128 scans per frame, 4 cm⁻¹             │
   │   · Total measurement time ~30 h            │
   ├─────────────────────────────────────────────┤
   │ Step 5: Automated Data Processing           │
   │   · siMPle / OpenSpecy automatic library search│
   │   · Candidate library: PE/PP/PS/PET/PA/PVC/PAN/PC/PMMA│
   │   · HQI > 0.85 considered positive          │
   │   · Output: particle position + size + polymer type│
   ├─────────────────────────────────────────────┤
   │ Step 6: Report                              │
   │   · Total particle count, size distribution  │
   │   · Polymer type distribution (pie chart)    │
   │   · Chemical class map (color-coded)         │
   └─────────────────────────────────────────────┘

5.2 siMPle and OpenSpecy: Open-Source Automated Analysis Tools

Automated processing of microplastic FPA data relies on specialized software [8][9]:

siMPle (standalone, developed by MFVPI Germany) [9]:

• Matlab-based, automatic library search + particle counting;
• Supports multiple FPA data formats (Bruker, Agilent, PerkinElmer);
• Provides reference library (PRIMPKE_A pristine polymer database);
• Suitable for research labs.

OpenSpecy (online, developed by Cowger et al. 2021) [8]:

• Web platform, no installation required;
• Supports uploading FPA data cubes;
• Built-in dynamic reference library (community-shared);
• Open-source (GitHub), supports secondary development;
• Suitable for cross-institution collaboration and teaching.

5.3 Typical Microplastic Research Results

Study 1: Microplastics in Chinese takeout containers (Zhou et al. 2022) [1]:

• Samples: 5 types of takeout containers (PP/PS/PET/PE/PLA);
• Method: FPA full-filter imaging;
• Result: 10,000–110,000 microplastics per container, particle size 11 μm–5 mm;
• Main polymer: consistent with container material (PP containers release PP particles);

Estimate: Each takeout meal ingestion contains 3–700 microplastic particles.

Study 2: Microplastic Verification in Portuguese Drinking Water (Cordeiro et al. 2025) [10]:

• Sample: 30 tap water samples from 5 cities;
• Method: μ-FTIR (with FPA) + validation methodology;
• Result: Average 0.7–14 particles per liter (safety threshold < 5 particles/L);
• Main polymers: PE, PP, PET (consistent with pipes and bottled water);
• Methodological contribution: Proposed LOD/LOQ validation framework for μ-FTIR detection of microplastics.

Study 3: Multi-city LDIR Study of Carbonated Beverages (Wang & Wang 2024) [11]:

• Sample: 36 carbonated beverages from 6 cities;
• Method: LDIR (Agilent 8700 laser infrared imaging: QCL + single-point detector high-speed scanning, not the same technology as FPA interferometric array imaging);
• Result: 5,000–80,000 particles per liter of beverage;
• Main polymers: PET (bottle body), PE (cap), PA (filter membrane).

🔗 Further Reading: ftir.fun functional group pages for common microplastic polymers:

• PE/PP: 2920/2850 cm⁻¹ CH₂ stretch → Alkyl C-H
• PS: 1495/1452 cm⁻¹ aromatic ring skeletal vibrations + 757/698 cm⁻¹ monosubstituted benzene out-of-plane bending → Aromatic
• PET: 1715 cm⁻¹ ester C=O + 1240 cm⁻¹ C-O → Ester
• PA: 1640/1540 cm⁻¹ amide I/II → Amide
• PVC: 1250 cm⁻¹ CH-Cl + 690 cm⁻¹ C-Cl → Chlorine
• PMMA: 1730 cm⁻¹ ester C=O + 1150 cm⁻¹ C-O → Methacrylate

6. Application Case 2: Tissue Section Chemical Imaging

6.1 Why Use FPA for Tissue Imaging?

Biological tissues are inherently chemically heterogeneous — a single section contains cells, extracellular matrix, lipid-rich regions, and protein aggregates. FPA chemical imaging can [12][13]:

• Label-free (no antibodies or fluorescent probes required);
• Non-destructive (does not damage the sample);
• Simultaneously acquire the distribution of multiple components such as proteins, lipids, nucleic acids, and glycogen;
• Directly register and compare with histological H&E staining images.

6.2 Tissue Sample Preparation

Sample preparation for tissue section FPA imaging is more complex than for microplastics [12][13]:

1. Cryosectioning: Fresh tissue → OCT embedding → cryosection at 8–10 μm;
2. Substrate selection: BaF₂ or CaF₂ windows (IR transparent, low background);
3. Avoid aldehyde fixation: Formalin introduces C-H interference;
4. Avoid moisture: Vacuum dry for 30 min before measurement;
5. Registration marks: Make physical scratches at the corners of the window for alignment with H&E images.

6.3 Key Chemical Information

Main biomolecular features of interest in tissue section FPA imaging [12][13][14]:

Table 6: Key biomolecular characteristic peaks in tissue FPA imaging

🔗 Further Reading: Core functional group pages involved in tissue chemical imaging:

• Protein secondary structure (Amide I): Protein, α-helix, β-sheet, random coil
• Lipids: Lipid, Fatty acid, Phospholipid
• Nucleic acids: Nucleic acid
• Glycogen and carbohydrates: Carbohydrate

6.4 Cancer Diagnosis Case

In 2015, the Baker group at the University of Manchester, UK, reviewed the application of FPA in cancer diagnosis in Chemical Society Reviews [13]:

Study design:

• Sample: Breast cancer tissue sections (n=120, including normal/ductal carcinoma in situ/invasive carcinoma);
• Method: FPA imaging (64×64, 5.5 μm pixel) + PCA/LDA classification;
• Key findings:
  • The Amide I center in cancerous tissue shifted from 1654 cm⁻¹ (predominantly α-helix) to 1632 cm⁻¹ (predominantly β-sheet), reflecting changes in protein secondary structure during malignant transformation;
  • The lipid/protein ratio (1740/1650) decreased in cancerous tissue, reflecting abnormal lipid metabolism;
  • The nucleic acid/protein ratio (1080/1650) increased, reflecting enlarged nuclei and increased DNA content in cancer cells;
• Diagnostic accuracy: LDA model achieved 92% on the validation set, comparable to pathologist interpretation.

Similar work has been carried out in lung cancer [14], colon cancer, prostate cancer, glioblastoma, and other cancers. FPA imaging is considered an important complement to future pathology [13].

6.5 Photizo: Open-Source Histopathology Tool

In 2022, the Dendrou Lab at the University of Oxford released Photizo — an open-source tool designed specifically for FTIR histopathology imaging [15]:

• Implemented in Python, supports FPA data cubes;
• Built-in preprocessing (baseline, normalization), PCA, K-means clustering, PLS-DA;
• Outputs chemical maps, classification maps, statistical metrics;
• Suitable for pathologists without programming background;
• GitHub: https://github.com/DendrouLab…

Photizo is a representative tool for the application of FPA in the biomedical field, lowering the research threshold [15].

7. Data Processing Challenges and Chemometrics

7.1 Big Data Challenges

The '4V' challenges of FPA data [3][7][8]:

• Volume: Single experiment GB–TB, annual accumulation PB;
• Velocity: Single FPA frame generates 25–100 MB, requiring real-time processing;
• Variety: Multi-vendor formats, multi-dimensional (spatial + spectral + temporal);
• Veracity: Low SNR, atmospheric interference, baseline drift require pixel-by-pixel correction.

7.2 Preprocessing Workflow

Typical preprocessing workflow for FPA data (each pixel processed independently) [3][8]:

   Raw data cube
        │
        ▼
   1. Atmospheric correction (CO₂ + water vapor)
      · Reference subtraction / EMSC / VaporFit
        │
        ▼

1. Baseline correction (per pixel)
   · Rubber band / Polynomial / ALS
   │
   ▼
2. ATR correction (if applicable)
   · Compensate wavelength-dependent penetration depth
   │
   ▼
3. Normalization
   · Area normalization / SNV / MSC
   │
   ▼
4. Smoothing / Derivation
   · Savitzky-Golay / Second derivative
   │
   ▼
5. Feature band selection
   · Reduce data dimensionality
   ```

Key tools: HyperSpy, SpectroChemPy, pybaselines (see Ep 53–54).

7.3 Chemometric Analysis

After preprocessing, the data cube enters chemometric analysis [3][6][8]:

1. Principal Component Analysis (PCA):

• Treat each pixel as a sample and all wavenumber channels as variables;
• Dimensionality reduction: 4000 wavenumbers → 5–10 principal components;
• Output: PCA score map (each principal component corresponds to a chemical map), loading spectra (reveals chemical information represented by principal components);
• Use: Unsupervised clustering, discovery of unknown chemical components.

2. Multivariate Curve Resolution (MCR-ALS):

• Decompose the data cube into pure component spectra + concentration maps;
• Output: Chemical spectrum and 2D distribution map for each component;
• Advantage: Can resolve overlapping peaks, separate mixed pixels.

3. Partial Least Squares Discriminant Analysis (PLS-DA):

• Supervised classification, requires labeled training set;
• Output: Class label for each pixel (e.g., "cancer" vs "normal");
• Use: Microplastic type discrimination, histopathological classification.

4. Deep Learning (CNN):

• Treat the data cube as a multi-channel image (each wavenumber as a channel);
• Use CNN to automatically learn chemical features;
• Frontier direction, suitable for large labeled datasets (see Ep 44).

7.4 OpenSpecy Workflow Example

OpenSpecy as an open-source microplastic analysis platform [8], its workflow:

   1. Upload FPA data cube (.dmd / .seq / .mat / ENVI)
        │
        ▼
   2. Automatic preprocessing (atmospheric correction, baseline correction, normalization)
        │
        ▼
   3. Particle identification (threshold segmentation → particle boundaries)
        │
        ▼
   4. Average spectrum per particle
        │
        ▼
   5. Library search (OpenSpecy dynamic reference library, containing 10000+ polymer spectra)
        │
        ▼
   6. Output report (particle count, particle size distribution, polymer type, chemical map)

OpenSpecy's openness makes it one of the "de facto standard" tools for microplastic research [8].

VIII. Limitations and Future Development of FPA Imaging

8.1 Current Limitations

1. SNR lower than single-point MCT: Limited photon flux per pixel, unfavorable for weak-absorbing samples [3];
2. Limited wavenumber range: Most FPAs cover 4000–900 cm⁻¹, far-infrared (< 900 cm⁻¹) requires special detectors [4][5];
3. Pixel crosstalk: Optical signal leakage between adjacent pixels, affecting true resolution [3];
4. Data explosion: Single experiment GB–TB level, storage and processing are engineering challenges [7][8];
5. High cost: 64×64 FPA detector priced at $50,000–150,000, 128×128 even more expensive;
6. Difficulty with synchrotron + FPA: FPA requires uniform illumination, while synchrotron beam is small and uneven in brightness, requiring special optical design [3].

8.2 Future Development Directions

1. Larger arrays: Commercialization of 256×256, 512×512 arrays, larger coverage per frame;
2. Room-temperature FPA: Based on Type-II superlattice or upgraded MCT, no liquid nitrogen needed, suitable for field use [4];
3. Fast-scan FPA: Frame rate 100+ fps, enabling time-resolved chemical imaging (see Ep 40);
4. AI automated analysis: Deep learning automatically identifies particles, classifies polymers, reducing manual intervention [8];
5. LDIR laser infrared imaging: Agilent 8700 etc. use quantum cascade lasers (QCL) + single-point detector + high-speed scanning, often faster than FPA (no traditional interferometer scanning), a strong contender in microplastics, but not an "equivalent technology" to FPA [11];
6. Cloud collaboration: Cloud platforms like OpenSpecy enable multi-institution sharing of data and models [8].

8.3 FPA vs LDIR: Emerging Competition

Table 7: FPA vs LDIR comparison (data source: Wang & Wang Heliyon 2024 [11])

LDIR is the main competitor to FPA in the microplastics field, each with its own advantages; market share will change dynamically over the next 5 years [11].

Summary of This Episode

Questions

1. Using a 64×64 FPA to measure a 47 mm filter requires approximately 19,000 frames to mosaic, total data volume ~470 GB. Design a data storage and processing plan: including raw data storage strategy, preprocessing pipeline, and automated report generation workflow. If switching to a 128×128 FPA, how would data volume, number of frames, and single-frame time change?

2. In tissue section FPA imaging, the Amide I center of cancer tissue redshifts from 1654 cm⁻¹ to 1632 cm⁻¹. Explain what protein secondary structure change this reflects and why this change is associated with malignant transformation. How can second derivatives further improve the resolution of Amide I sub-peaks?

3. The SNR of a single FPA pixel is lower than that of a single-point MCT detector because of 'low single-pixel luminous flux'. Please explain from an optical principle: Why does a single FPA pixel receive less infrared energy than a single-point MCT detector? What impact does this have on the limit of detection (LOD) for microplastic analysis?

4. Compare the advantages and disadvantages of FPA and LDIR in microplastic analysis. If you are the director of an environmental monitoring station with a budget of $300,000 and need to process 1000 water samples per year, which technology would you choose? Why?

5. Design an FPA experiment: Study the phase separation structure of a 5 mm × 5 mm polymer blend (PE/PP 70:30) thin section. Please describe sample preparation, FPA parameters (array, objective, pixel size), scan time estimation, data processing workflow (including how to distinguish PE from PP).

References

[1] Zhou X, Wang J, Ren J. "Analysis of Microplastics in Takeaway Food Containers in China Using FPA-FTIR Whole Filter Analysis." Molecules, 2022, 27(9): 2646. DOI:10.3390/molecules27092646

[2] Griffiths P R, de Haseth J A. Fourier Transform Infrared Spectrometry. 2nd ed. Wiley, 2007. ISBN: 978-0-471-19404-0.

[3] Nasse M J, Walsh M J, Mattson E C, et al. "High-Resolution Fourier-Transform Infrared Chemical Imaging with Multiple Synchrotron Beams." Nature Methods, 2011, 8(5): 413–416. DOI:10.1038/nmeth.1580

[4] Bruker Optik GmbH. Hyperion 3000 FTIR Microscope User Manual. 2020.
https://www.bruker.com/en/pro…

[5] Agilent Technologies. 620 FTIR Imaging System Specifications. 2021.
https://www.agilent.com/en/pr…

[6] Bhargava R. "Infrared Spectroscopic Imaging: The Next Generation." Applied Spectroscopy, 2013, 66(10): 1091–1120. DOI:10.1366/12-06701

[7] Primpke S, Fischer M, Lorenz C, et al. "Comparison of Imaging and FTIR Microscopy Applied for the Analysis of Microplastics in Environmental Samples." Applied Spectroscopy, 2020, 74(9): 1185–1197. DOI:10.1177/0003702820922969

[8] Cowger W, Steinmetz Z, Gray A, et al. "Microplastic Spectral Classification Needs Open Source Tools: Introducing OpenSpecy." Analytical Chemistry, 2021, 93(32): 11184–11193. DOI:10.1021/acs.analchem.1c01349

[9] Primpke S, Wirth M, Lorenz C, Gerdts G. "Reference Database Design for the Automated Analysis of Microplastic Samples Based on Fourier Transform Infrared (FTIR) Microscopy." Analytical and Bioanalytical Chemistry, 2018, 410(21): 5131–5141. DOI:10.1007/s00216-018-1156-x

[10] Cordeiro R D M et al. "Validation of an FT-IR Microscopy Method for the Monitorization of Microplastics in Water for Human Consumption in Portugal." Environmental Science and Pollution Research, 2025, 32(28): 16823–16844. DOI:10.1007/s11356-024-33966-8

[11] Wang Y, Wang Y. "Assessing Microplastic Contamination in Soda Beverages Using LDIR." Heliyon, 2024, 10(12): e32805. DOI:10.1016/j.heliyon.2024.e32805

[12] Lasch P, Naumann D. "Spatial Resolution in Infrared Microspectroscopic Imaging of Tissues." Biochimica et Biophysica Acta — Biomembranes, 2006, 1758(7): 814–829. DOI:10.1016/j.bbamem.2006.06.008

[13] Baker M J, Trevisan J, Bassan P, et al. "Using Fourier Transform IR Spectroscopy to Analyze Biological Materials." Nature Protocols, 2014, 9(8): 1771–1791. DOI:10.1038/nprot.2014.110

[14] Hughes C, Baker M J. "Can Mid-Infrared Biomedical Spectroscopy of Healthy, Diseased and Cancerous Tissues Contribute to Treatment?." Analyst, 2016, 141(2): 467–475. DOI:10.1039/C5AN01858G

[15] Pember B, Smith K, Dendrou C A. "Photizo: An Open-Source Computational Tool for FTIR Histopathology." Bioinformatics, 2022, 38(13): 3490–3491. DOI:10.1093/bioinformatics/btac346

[16] ftir.fun Alkyl C-H functional group page. https://ftir.fun/ir/group/alk…

[17] ftir.fun Ester functional group page. https://ftir.fun/ir/group/est…

[18] ftir.fun Amide functional group page. https://ftir.fun/ir/group/ami…

[19] ftir.fun Protein functional group page. https://ftir.fun/ir/group/pro…

[20] ftir.fun Protein α-Helix Functional Group Page. https://ftir.fun/ir/group/pro…

[21] ftir.fun Protein β-Sheet Functional Group Page. https://ftir.fun/ir/group/pro…

[22] ftir.fun Protein Random Coil Functional Group Page. https://ftir.fun/ir/group/pro…

[23] ftir.fun Lipid Functional Group Page. https://ftir.fun/ir/group/lip…

[24] ftir.fun Phospholipid Functional Group Page. https://ftir.fun/ir/group/pho…

[25] ftir.fun Nucleic Acid Functional Group Page. https://ftir.fun/ir/group/nuc…

[26] ftir.fun Carbohydrate Functional Group Page. https://ftir.fun/ir/group/car…

[27] ftir.fun Aromatic Ring Functional Group Page. https://ftir.fun/ir/group/aro…

[28] ftir.fun Chlorine Functional Group Page. https://ftir.fun/ir/group/chl…

[29] ftir.fun Methacrylate Functional Group Page. https://ftir.fun/ir/group/met…

Next Episode Preview: Ep 38 — Synchrotron Infrared Source: Ultra-bright, Ultra-resolution
In this episode, we saw how FPA breaks through the throughput bottleneck with "parallel measurement." The next episode will focus on another "performance amplifier"—the synchrotron infrared source. We will explain the physical characteristics of synchrotron radiation (broad spectrum, high brightness, high collimation), compare its 10–100× SNR improvement over conventional blackbody sources, introduce major synchrotron infrared beamlines worldwide (SOLEIL, ALS, SSRF), and demonstrate how synchrotron radiation pushes μ-FTIR to the diffraction limit through single-cell imaging and nanomaterial characterization cases.

This article is licensed under CC BY-NC-SA 4.0. Images are from public domain or attributed online resources, copyrights belong to respective owners.