Owing to the minimal invasiveness, cytology is an indispensable technique in the routine pathology. However, traditional cytology only enables the low sensitivity (50%-60%) and high time-consuming for the diagnosis. Our previous study demonstrated that stimulated Raman molecular cytology (SRMC), which is label-free, faster, and noninvasive, provides additional composition information leading to higher diagnostic accuracy around 85%. However, current AI-assisted SRMC generally involves cell segmentation and feature extraction steps, which may involve issues of the artifacts. Recently, various methods for global feature analysis, such as Transformer and CNN, are capable of preserving both global and local information. Therefore, we propose an end-to-end Transformer hybrid model combining the advantages of both Transformer and CNN to analyze stimulated Raman cytology images for accurate and rapid peritoneal metastasis (PM) diagnosis of gastric cancer (GC). The Transformer hybrid model can enhance the Transformer’s global modeling ability simultaneously with the local guidance from CNN features. To evaluate the performance of this Transformer method, we collected 816 stimulated Raman cytology images from 80 locally advanced gastric cancer patients, with 36 PM positive and 44 PM negative. The Transformer method could reach 88.89% sensitivity, 86.36% specificity, and an AUC of 0.903 with leave-one-out cross-validation for 80 patients. Compared with traditional cytology, the false negative rate of our label-free stimulated Raman cytology reduces by about 30-50%. Together, our Transformer approach demonstrates the potential for accurate and rapid PM diagnosis based on exfoliated cytology.
Current Hyperspectral stimulated Raman scattering (hsSRS) data analysis methods face challenges when it comes to rapidly and reliably quantifying different lipid subtypes, and cannot fully leverage the information in hsSRS data. Here, we present a rapid and reliable quantitative algorithm for quantitative analysis that fully extracts chemical information by using adaptive selection of Lorentzian basis functions to fit the spectra in hsSRS data in bulk. We demonstrated that, by utilizing the ratio relationships between fitted bands, quantitative comparisons of specific lipid subtypes can be achieved. Moreover, we applied our method for the quantitative analysis of lipid composition in lipid droplets based on hsSRS data of liver cancer tissues and confirmed our method has a better fitting effect and a faster solving speed compared to MCR. This suggests that our method has the potential for great utility in the quantitative analysis of hsSRS imaging data for biomedical specimens.
Heterogenous zonation plays a crucial role in liver physiology and pathology. Extracellular collagen fiber deposition and intracellular lipid accumulation are hallmarks of many liver diseases, but their zonal distribution and pathological significance remains unknown. Here, we established a label-free multimodal nonlinear optical (NLO) microscopy that integrated stimulated Raman scattering (SRS) imaging and compositional analysis of neutral lipids, second harmonic generation (SHG) imaging of collagen fibers, and two-photon excited fluorescence (TPEF) imaging of auto-fluorescent metabolites to study liver zonation in situ. By analyzing the liver tissues obtained from carbon tetrachloride (CCl4) induced mouse centrilobular fibrosis model, our imaging data revealed that the amount of collagen fibers increased rapidly in central vein (CV) area, while significantly slower in portal vein (PV) area. Along with the collagen fibers deposition, the accumulation of triglycerides (TGs) was also significantly increased in CV area, but almost absent in PV area. Within the total TGs, although the amount of unsaturated TGs increased significantly in CV area, the lipid unsaturation degree, defined as the number of C=C bands on each fatty acid chain, significantly decreased in the late-stage compared to the early-stage fibrosis, which was likely due to the change of stearoyl-coenzyme A desaturase-1 (SCD1) expression. Meanwhile, lipofuscin and malondialdehyde, products of lipid peroxidation, were found to be gradually accumulated in CV area as fibrosis progressed. Interestingly, the heterogeneous distribution of collagen fibers and lipids were also found in fibrotic livers in patients. Our study by label-free multimodal NLO microscopy discovered the liver zonation of TG and collagen fiber, which could serve as valuable biomarkers for assessment of liver fibrosis progression.
The survival rate for renal cancer patients is closely related to the surgical margin status. Thus, accurate and rapid detection of renal cancer is needed. Here, we integrated photoacoustic tomography (PAT) with ultrasound imaging in a single system, which achieved tissue imaging depth about 3 mm and imaging speed about 3.5 cm2/min. We used the wavelength at 1064 nm and 1197 nm to map both blood and lipid distribution in 16 normal and 17 clear cell renal cell carcinoma (ccRCC) tissues, collected from nephrectomy. Our results indicated that the photoacoustic signal from lipids, but not blood, was significantly higher in ccRCC tissues than that in normal tissues. Moreover, based on the quantification of lipid area ratio, we were able to differentiate normal and ccRCC with 100% sensitivity, 80% specificity, and area under receiver operating characteristic curve of 0.95. Our findings show promise of using multimodal PAT for intraoperative ccRCC detection.
Altered lipid metabolism is increasingly recognized as a signature of cancer cells. Enabled by label-free spectroscopic imaging, we performed quantitative analysis of lipogenesis at single-cell level in human clear cell renal cell carcinoma (ccRCC), which accounts for about 90% kidney cancers. Our hyperspectral stimulated Raman scattering (SRS) imaging data revealed an aberrant accumulation of lipid droplets in human clear cell renal cell carcinoma (ccRCC), but no detectable lipid droplets in normal or benign kidney tissues. We also found that such lipid accumulation was significantly higher in low grade (Furhman Grade≤2) ccRCC compared that in high grade (Furhman Grade≥3) ccRCC, and was correlated well with the prognosis of ccRCC. Moreover, cholesteryl ester is the dominant form of lipids accumulated in ccRCC. Besides, the unsaturation level of lipids was significantly higher in high grade ccRCC compared to low grade ccRCC. Furthermore, depletion of cholesteryl ester storage significantly reduced cancer proliferation, impaired cancer invasion capability, and suppressed tumor growth and metastasis in mouse xenograft and orthotopic models, with negligible toxicity. These findings herald the potential of using lipid accumulation as a marker for diagnosis of human ccRCC and open a new way of treating aggressive human ccRCC by targeting the altered lipid metabolism.
Due to the subject nature of histopathology, there is a significant inter-observer discordance for the differentiation between low-risk prostate cancer (Gleason score ≤ 6), which can be left without treatment, and high-risk prostate cancer (Gleason score >6), which requires active treatment. Our previous study using Raman spectromicroscopy reveals that cholesteryl ester accumulation underlies human prostate cancer aggressiveness. However, Raman spectromicroscopy could only provide compositional information of certain lipid droplets of interest, which overlooked cell-to-cell variation and hindered translation to accurate automated diagnosis. Here, we demonstrated quantitative mapping of cholesteryl ester molar percentage in human prostate cancer tissues using hyperspectral stimulated Raman scattering microscopy that renders compositional information for every pixel in the image. Specifically, hundreds of SRS images at Raman shift between 2800~3000 cm-1 were taken, and multivariate curve resolution algorism was used to retrieve concentration images of lipid, lipofuscin, and protein. We found that the height ratio between the prominent cholesterol band at 2870 cm-1 and the CH2 stretching band at 2850 cm-1 was proportional to the molar percentage of cholesteryl ester present in the total lipids. Based on the calibration curve, we were able to quantitatively map cholesteryl ester level in intact prostate cancer tissues. Our data showed that not only the amount of cholesteryl ester-rich lipid droplets, but also the CE molar percentage, was significantly greater in prostate cancer tissues with Gleason score > 6 compared to the ones with Gleason score ≤ 6. Our study offers an opportunity towards more accurate prostate cancer diagnosis.
Most prostate cancers (PCa) are slowly growing, and only the aggressive ones require early diagnosis and effective treatment. The current standard for PCa diagnosis remains histopathology. Nonetheless, for the differentiation between Gleason score 6 (low-risk PCa), which can be left without treatment, and Gleason score 7 (high-risk PCa), which requires active treatment, the inter-observer discordance can be up to 40%. Our previous study reveals that cholesteryl ester (CE) accumulation induced by PI3K/AKT activation underlies human PCa aggressiveness. However, Raman spectromicroscopy used in this study could only provide compositional information of certain lipid droplets (LDs) selected by the observer, which overlooked cell-to-cell variation and hindered translation to accurate automated diagnosis. Here, we demonstrated quantitative mapping of CE level in human prostate tissues using hyperspectral stimulated Raman scattering (SRS) microscopy that renders compositional information for every pixel in the image. Specifically, hundreds of SRS images at Raman shift between 1620-1800 cm-1 were taken, and multivariate curve resolution algorism was used to retrieve concentration images of acyl C=C bond, sterol C=C bond, and ester C=O bond. Given that the ratio between images of sterol C=C and ester C=O (sterol C=C/C=O) is nonlinearly proportional to CE percentage out of total lipid, we were able to quantitatively map CE level. Our data showed that CE level was significantly greater in high Gleason grade compared to low Gleason grade, and could be a factor that significantly contributed to cancer recurrence. Our study provides an opportunity towards more accurate PCa diagnosis and prediction of aggressiveness.
We demonstrate nonlinear vibrational imaging of isolated Raman bands by detecting femtosecond pulse stimulated
Raman loss. Femtosecond pulse excitation produces a stimulated Raman loss signal that is 12 times larger than what
picosecond pulse excitation produces. The strong signal allowed real-time, bond-selective imaging of deuterated palmitic
acid-d31 inside live cells, and 3D sectioning of fat storage in live C. elegans. With the high peak power provided by
femtosecond pulses, this system is highly compatible with other nonlinear optical modalities such as two-photon excited
fluorescence. With most of the excitation power contributed by the Stokes beam in the 1.0 - 1.2 μm wavelength range,
photodamage of biological samples was not observed.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.