Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Pipeline of the reconstruction network. The reconstruction network comprises three stages. In the first stage, it takes as input a low spatial but high spectral resolution hyperspectral image (LR-HSI) from, for instance, a snapshot camera and a high spatial but low spectral resolution RGB image (HR-RGB) from a regular color camera. These two inputs are subsequently fused to create a preliminary RGB-HSI image. To achieve this, the LR-HSI is upsampled to a dimension of 448 × 448. Following the fusion process, the RGB-HSI is then processed through the second and third stages. In these stages, the network is trained to independently reconstruct the spatial and spectral information of the image using the reference image as ground-truth. This ultimately produces a hyperspectral image with a high spatial and spectral resolution (HR-HSI).

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques:

Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Blurring experiment: example of a lumpectomy specimen after surgery. Pseudocolor images (wavelength bands at 650, 532, 473 nm) are derived from the hyperspectral images. From left to right: Gaussian blurred LR-HSI, upsampled LR-HSI, reconstructed HR-HSI, and reference HR-HSI. The LR-HSI images are blurred with Gaussian filters of different kernel sizes, representing snapshot HSI images in a surgical setting. The upsampled and reconstructed images show the result when, respectively, upsampling and reconstructing the blurred images to high-resolution images. From top to bottom: Gaussian filters with kernel sizes of 5 × 5 , 15 × 15 , 25 × 25 and 35 × 35 .

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques: Derivative Assay

Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Noise experiment: example of a lumpectomy specimen after surgery. Gray color images are derived from the hyperspectral images with a wavelength band at 980 nm. From left to right: Noise LR-HSI, reconstructed HR-HSI, and reference HR-HSI. Noise was introduced into the spectra of the LR-HSI images to simulate the reduced spectral sensitivity of the camera sensor within the 980–1000 nm range. From top to bottom: Noise with a variance of 0.01, 0.03, 0.05, and 0.07.

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques: Derivative Assay

Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Dead pixels experiment. Pseudocolor images (wavelength bands at 650, 532, 473 nm) are derived from the hyperspectral images. Top row: lumpectomy specimen containing 12 dead pixels (black) on its resection surface, and associated image after reconstruction. Bottom row: enlarged view (magnification∼10×) of two delineated dead pixels highlighted in yellow.

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques: Derivative Assay

Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Comparison of reflectance spectra from two dead pixels that were restored in the reconstructed hyperspectral image. The upper pixel in the magnified image of corresponds to ( a ) and the lower pixel to ( b ). The shaded areas in gray represent the wavelengths at which the camera sensor exhibits reduced spectral sensitivity.

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques: Comparison

Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Specular reflection experiment. Gray color images derived from the hyperspectral images with a wavelength band at 705 nm. Top row: LR-HSI of lumpectomy specimen containing four regions of specular reflection in white, and associated result after reconstruction. Bottom row: enlarged view (magnification∼10×) of four regions with specular reflection.

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques: Derivative Assay

Journal: Sensors (Basel, Switzerland)

Article Title: Spatial and Spectral Reconstruction of Breast Lumpectomy Hyperspectral Images

doi: 10.3390/s24051567

Figure Lengend Snippet: Comparison of reflectance spectra from four specular reflection regions that were restored in the reconstructed hyperspectral images. A comparison is made between the center pixels of the top-left ( a ), top-right ( b ), bottom-left ( c ), and bottom-right region ( d ) with a size of, respectively, 4, 9, 16, and 25 pixels. The regions are depicted in the magnified image of . In each spectrum, the gray shaded area highlights the wavelength range associated with the specular reflection (675–705 nm).

Article Snippet: The images were acquired using a line-scanning hyperspectral camera (Specim, Spectral Imaging Ltd., Oulu, Finland, PFD-CL-65-V10E, linear CMOS sensor), including a spectral range between ∼400–1000 nm (384 bands), a spectral resolution of 3 nm, and spatial resolution of 0.16 mm per pixel.

Techniques: Comparison