1.

Introduction

The emergence of minimally invasive diagnostic and interventional procedures has increased the demand for real-time guidance for a variety of oncology applications that rely on accurate instrument positioning and procedural feedback. To this aim, optical spectroscopy has been investigated as a tool to increase the accuracy and efficacy of clinical procedures.

As spectroscopy measurements can be conducted through thin needles (28 G; 0.36 mm),¹ there is a strong rationale to perform optical spectroscopy measurements at the tip of a needle for rapid tissue characterization. This can be used for biopsy guidance by providing more selective tissue acquisition² and increasing the diagnostic yield and biopsy quality.³ The development of advanced needles for needle-based thermal therapies, such as radiofrequency ablation, microwave ablation, and cryoablation in the liver, has offered minimally invasive therapies to patients that were previously untreatable. The success of these treatments is highly dependent on the accurate placement of the ablation needles, therefore, tissue characterization at the tip of the device would be of significant value.^4–6 The functionality of optical spectroscopy added to a surgical instrument could help surgeons to find the optimal resection plane and ensure completeness of tumor removal, as the recognition of the tumor tissue during surgery is often challenging,⁷ especially in minimally invasive procedures.

Diffuse reflectance (DR) spectroscopy is a widely used technique to estimate the optical properties of tissue where tissue is illuminated with a selected spectral band of light. The light is either scattered or absorbed by the tissue, depending on the specific composition of the tissue. Subsequent analysis of the tissue’s spectral response provides specific quantitative morphologic, biochemical, and functional information, thereby enabling tissue discrimination and potentially improving diagnostic capability.

Several preclinical studies were performed in the last decade showing the potential of DR spectroscopy to discriminate liver tumors and surrounding liver tissue.^1,8–10 Nachabé et al.^8,9 investigated differences between healthy tissue and metastatic tumor from ex vivo human liver specimens using diffuse optical spectroscopy with a fiber optic probe. It was shown that bile should be included when analyzing DR liver data because of its presence in the liver bile ducts and its strong optical signature.⁸ In a preclinical (ex vivo) study by Evers et al.,⁹ bile content was found to be significantly higher in normal liver tissue compared to colorectal liver metastases. In addition to bile, the main chromophore in liver tissue dominating the absorption in the visible spectrum (400 to 750 nm) is hemoglobin (oxygenated and deoxygenated).^8,11 Water and fat dominate in the near-infrared range ($>800\mathrm{nm}$).^8,12,13 The main scattering parameters in liver tissue are the reduced scattering at 800 nm and the Mie-to-total scattering fraction, in which the total scattering of tissue is assumed to be defined by Mie and Rayleigh scattering.^8,9 In previous ex vivo studies on 14 and 24 resection specimen on average, a lower reduced scattering (at 800 nm) was found in colorectal liver metastases in comparison with healthy liver tissue.^8,9

A considerable challenge in DR spectroscopy-based tissue characterization during clinical procedures (e.g., needle guidance and surgical guidance) is the inevitable presence of blood. Earlier it was shown that extending the measurement into the near-infrared region up to 1600 nm, where blood has no significant absorption features, helps to overcome the effect of dominant absorption by excessive amounts of hemoglobin in the visible wavelength region (400 to 700 nm) that can occur due to bleeding. Whether bile, which is strong absorber in the visible wavelength range, can still be measured accurately under specific operating constraints in the clinic (e.g., when considerable amounts of blood are encountered) has not been elucidated.

Histology remains the gold-standard for diagnosis and staging in oncology, making it in essence a ground truth for new technologies attempting to measure and quantify these pathologies. However, clinical validation studies are often confounded by imperfect spatial correlation of the optical reading and the tissue sample removed for histopathological analysis.^14,15 An approach to address discrepancies between spectral data and histology is to integrate fiber-optics into standard tissue-sampling tools such as a core biopsy needle, thereby linking spectroscopy and biopsy functionality in a single instrument.² This feature guarantees a near-perfect match between optically sampled and histologically analyzed tissue.

The objective of the present study is to assess the in vivo tissue discriminating abilities of DR spectroscopy in a series of image-guided liver biopsy procedures using a core biopsy needle with integrated optical fibers.

2.

Materials and Methods

3.

Results

Of the 21 patients, the median age was 65.9 yr (range 48.1 to 83.7 yr). Eleven participants (52%) were women. The median tumor size was 27 mm (range 11 to 71 mm). Histopathological examination of the targeted tissue revealed 20 malignancies, of which 17 were classified as metastasis from colorectal adenocarcinomas (Table 1). One tumor was identified as a metastasis from a breast tumor and two lesions were metastases of a neuroendocrine tumor. Two biopsies yielded only nondiagnostic material (subjects 9 and 10). In two patients (subjects 5 and 14), there was a mismatch between final DR measurement position and biopsy location due to needle movement after the final DR measurements. For these patients, DR recordings were only linked to the imaging and not with tissue classification based on histology.

Table 1

Demographics and individual data for patients (n=21).

3.1.

Cohort Analysis

Figure 1 illustrates how DR tissue characterization was performed using the FOBN. Added real-time spectral measurements did not interfere with the standard biopsy procedure. When the collective data ($n=333$ DR spectra) were compared (Fig. 2), significant compositional differences ($p<0.01$) between tumor tissue and liver tissue were noted for blood, ${\mathrm{stO}}_2$, bile, and ${\mu }_{\mathrm{s}}^{\prime }(800)$. Area under the corresponding ROC curves was used to measure the ability of various DR spectroscopy parameters to discriminate tumor tissue from normal liver tissue (Fig. 2). Blood content, ${\mathrm{stO}}_2$, and ${\mu }_{\mathrm{s}}^{\prime }(800)$ demonstrated an AUC of 0.659 to 0.708, indicative of poor to moderate discrimination. Classification based on bile content allowed more accurate identification of tumor tissue (ROC–AUC value: 0.924). The optimum cutoff value of the ROC curve corresponded with a sensitivity of 0.91% and a specificity of 0.88%.

Fig. 1

Example of added quantitative spectral functionality during routine liver biopsy (first patient 12 in Table 1). (a) Positioning of the FOBN based on CT-fluoroscopic imaging in liver tissue, at the border of the target lesion and in the target lesion prior to biopsy. (b) Coregistered DR measurements (blue dotted line) and corresponding fit curves (red lines). (c) Data for blood, ${\mathrm{stO}}_2$, bile, and ${\mu }_{\mathrm{s}}^{\prime }(800)$ represent mean values ± standard deviations. N, normal liver tissue; B, tumor border; T, tumor; and FOBN, fiber-optic biopsy needle.

Fig. 2

DR spectroscopy parameter quantification. Bar graphs show the values for (a) blood, (c) ${\mathrm{stO}}_2$, (e) bile, and (g) ${\mu }_{\mathrm{s}}^{\prime }(800)$ as measured in liver tissue (normal), at the tumor border (border) and in the target lesion (tumor). Values are given as $\text{mean values}\pm \text{standard errors}$, adjusted for repeated measurements. *$p<0.01$. NS: not significant. Corresponding receiver operating characteristics (ROC) curves (b, d, f, h) indicate the ability to distinguish tumor tissue from liver tissue. AUC: area under ROC curve.

3.2.

Individual Patient Analysis

Discrimination between various tissues relies on measurement of the absolute or relative differences in intensity or optical contrasts between the tissue types of interest. Sources of variation, such as interpatient variability, may outweigh the difference between the tissue types of interest and lead to hampered diagnostic performance.^9,17,18 Relative changes in DR parameters that occur in a particular insertion may help to define more effective detection criteria that are less sensitive to interpatient variation. Therefore, we determined the relative contrast between tumor and normal liver tissue based on the spectroscopically derived values for bile content using each patient’s normal liver tissue as an internal reference (Fig. 3; Table 1). In 22 out of the 24 biopsy procedures, tumor tissue could be correctly identified based on a decrease in bile content. In the two remaining cases (subject 9 and subject 10), the bile content measured just before biopsy indicated that FOBN was not in contact with the tumor tissue (Fig. 4). In both cases, the histopathological analysis of the biopsy samples confirmed that the targeted lesion was missed. If this feedback had been used, a corrective manipulation of the needle could have increased the chance of an adequate biopsy.

Fig. 3

Relative changes in bile content determined for $n=24$ insertions. The relative bile content was defined as the bile content in tumor relative to the average value measured in normal liver tissue. Values are given as the $\text{mean}\pm \text{standard deviations}$. Dotted line: relative bile $\text{content}=1$.

Fig. 4

(a) Although CT-fluoroscopic imaging in subject 9 and subject 10 suggests that the tissue biopsy was taken from the target lesions, the tissue samples proved to be nondiagnostic. The samples contained only normal liver tissue, indicating that the targeted tissue was missed. No substantial change in bile content and other DR spectroscopy parameters was seen. These examples underline the importance of real-time measurements and data analysis in order to identify the transition of needle placement in a tumor based on the changes in the derived parameters. (b) Data for blood, ${\mathrm{stO}}_2$, and bile represent $\text{mean values}\pm \text{standard deviations}$. N, normal lung tissue; T, tumor; and FOBN, fiber-optic biopsy needle.

3.3.

Real-Time Tissue Characterization

In three patients (subjects 19 to 21), spectral measurements were acquired continuously during needle insertion, as exemplified in Fig. 5. While the needle tip was progressed toward the tumor, CT-fluoroscopic imaging provided the actual location of the needle tip. A decrease in fat content was observed when the tip of the needle was inserted from subcutaneous fat/muscle into the liver. The ${\mathrm{stO}}_2$ level strongly decreased near the tumor border and in the tumor. Moreover, a clear decrease in bile content was observed once the needle reached the tumor.

Fig. 5

Real-time tissue characterization and biopsy with the FOBN in an individual. (a) CT-fluoroscopy images during insertion of the FOBN (images I through VI). (b) DR spectroscopy parameters were extracted from the optical data along the needle path. Note that the spectral measurements were started while the FOBN was positioned in subcutaneous fat or muscle. FOBN, fiber-optic biopsy needle.

4.

Discussion

The development of advanced fiber-optic clinical tools could make a significant contribution to diagnosis and treatment monitoring in cancer patients.

The goal of this study was to evaluate the diagnostic potential of DR spectroscopy during image-guided liver biopsy procedures. For this, a core biopsy needle with integrated optical fibers was used to link spectral measurements and biopsy functionality in a single instrument. The use of such an instrument represents a major step forward for clinical evaluation of spectral tissue sensing techniques by greatly increasing the spatial correlation of physical biopsies with spectral measurement spots and simplifying study procedures

Diagnostic performance of our system was tested during 24 image-guided percutaneous liver biopsy procedures. Tissue blood content, bile content, ${\mathrm{stO}}_2$, and ${\mu }_{\mathrm{s}}^{\prime }(800)$ showed statistically significant differences between tumor tissue and surrounding liver tissue. These findings are consistent with previous preclinical studies.^8–10

We found that bile content allows accurate discrimination between normal liver tissue and tumor tissue (ROC AUC: 0.924). Relative increases in bile content enabled identification of tumor tissue in all 22 of 24 the clinical cases. In the remaining two cases, DR spectroscopy correctly predicted that the biopsy specimen did not contain tumor tissue. In this series, this corresponds with an overall diagnostic performance of 100%

Normal liver tissue mainly consists of hepatocytes, which are cells that are arranged as very thin plates separated by fine vascular sinusoids where blood flows. Bile produced by liver cells drains into microscopic canals known as bile canaliculi. The lower bile content in secondary liver cancers can be attributed to the loss of the native tissue structure and associated bile perfusion. Whereas bile content is directly linked to tissue structure, the measured blood content is strongly susceptible to bleeding at the tip of the fiber-optic needle.² Consequently, estimates for blood content and associated oxygenation levels do not necessarily reflect the true physiologic state of the measured tissue, thus rendering them less suitable for in vivo tissue discrimination.

Regarding ${\mu }_{\mathrm{s}}^{\prime }(800)$, our results agree with the general reported findings from other (ex vivo) human DR spectroscopy studies.^8,9 Liver parenchyma was found to have higher ${\mu }_{\mathrm{s}}^{\prime }(800)$ as compared to tumor. This suggests that liver tissue has a larger density of small particles than tumors. This observation collaborates with the fact that healthy liver tissue is rich in densely packed small hepatocytes compared to colorectal liver metastases in the liver. The tissue that is present at the interface between tumor and nontumor tissue strongly depends on the histological growth pattern of such a metastasis.²⁰ For example, metastases that develop according to a “desmoplastic” or “pushing” growth pattern are separated from surrounding liver parenchyma by a rim of desmoplastic stroma or inflammatory infiltrate, whereas in a “replacement” growth pattern, there is generally no rim or capsule around the metastasis. It is unclear whether the lower values for ${\mu }_{\mathrm{s}}^{\prime }(800)$ that were measured with DR spectroscopy at the tumor borders (Fig. 2) are caused either by specific histological characteristics of the tumor or by limitations of the analytical model, which does not take into account multiple layers of tissue.

The information provided by DR spectroscopy, such as mapping of bile levels along the needle path, can be of great relevance in the case of biopsy procedures or other procedures that rely on accurate needle placement. Continuous DR measurements, as performed during this study, may therefore provide additional information with respect to the variation in tissue composition across the tumor as well as identifying the boundaries between normal liver tissue and tumor tissue.

The current experimental system was designed for clinical studies and has proven to be compatible with the existing clinical workflow of percutaneous liver biopsy procedures. Further clinical validation of the outlined approach in a prospective setting or (subsequent) clinical implementation would require that the spectral information is translated into comprehensible and reliable real-time diagnostic information. The optical fiber geometry will be an important consideration for the design of a future fiber-optic tool. For example, probing volume and depth strongly depend on the tissue’s optical properties and the source–detector fiber separation. As the source–detector separation increases, the probing depth and volume increases, which increases the sensitivity of the probe to deeper tissues. However, the ability to detect small amounts of abnormal tissue decreases with a larger probed volume (partial volume effect). The challenge will be to develop a system with a sufficient probing depth and sufficient sensitivity to detect tumor margins or progressive infiltration of tumor into liver parenchyma. Thus, decisions about technical details and the algorithms used for extracting tissue optical properties will be dictated largely by the application needs. This will require further data collection and validation studies.

During biopsy procedures, physicians try to avoid multiple passages to prevent tumor seeding, pain, bleeding, and so on. Continuous real-time DRS measurements, as performed during this study (Fig. 5), can be used to record a profile along the needle path and yield valuable data with respect to the variation in tissue composition across the tumor as well as identifying the boundaries between tissues. Because DR spectroscopy is performed from tissues that are close to the fiber-optic probe tip (1 to 2 mm), we expect that spectral information will ultimately be used complementary to image guidance (e.g., fluoroscopy or ultrasound imaging). By providing crucial information during biopsy needle insertion and just before the tissue sample is taken, the outlined approach may help to reduce the number of false-negative biopsies with minimal impact on routine clinical workflow.

Earlier, we found that the reliability (i.e., confidence intervals) of various DR spectroscopy parameters is not affected by the amount of blood at the fiber-optic probe tip.² However, absolute parameter values will depend to some extent on the amount of blood encountered, as blood close to the tip is part of the probed “tissue” volume. The measured blood content at the needle tip—which is accurately quantified—could be provided as feedback to the physician, thereby allowing necessary adjustments of the fiber-optic probe to reduce the effect of blood. Moreover, in the case of surgical guidance, DR spectroscopy is likely to be used in combination with a cauterization instrument. For such an application, an instrument has to be developed that functions reliably in the complex surgical setting. If the instrument is expected to alert the surgeon that tumor tissue is detected, the system must be able to identify any normal or tumor tissue type that could potentially be encountered by the operator during an intervention under all foreseeable operating conditions. Furthermore, surgeons often use electrocauterization to cut through soft tissue and to seal off blood vessels that are bleeding during surgery. If DR spectroscopy is used in combination with a cauterization instrument, it will be necessary to understand the effects of heat-induced tissue alterations, such as tissue ablation, carbonization, water vaporization, and thermal denaturation of proteins. The need for such smart surgical instruments is increasing as the standard of care liver surgery tries to spare as much healthy liver tissue as possible, meaning that the number of local and segmental resections increases. Especially with these types of resections, tumor margins are more difficult to judge, especially during laparoscopic liver resection.

5.

Conclusion

In summary, in this study, we provided an important validation step for DR spectroscopy-based tissue characterization in liver. We found that bile content can be used as the primary discriminator for transitioning from normal liver tissue to tumor. Given the feasibility of the outlined approach, it is also conceivable to make integrated fiber-optic tools for other oncological procedures that rely on accurate instrument positioning.