Towards machine learning-based quantitative hyperspectral image guidance for brain tumor resection.

BACKGROUND: Complete resection of malignant gliomas is hampered by the difficulty in distinguishing tumor cells at the infiltration zone. Fluorescence guidance with 5-ALA assists in reaching this goal. Using hyperspectral imaging, previous work characterized five fluorophores' emission spectra in most human brain tumors. METHODS: In this paper, the effectiveness of these five spectra was explored for different tumor and tissue classification tasks in 184 patients (891 hyperspectral measurements) harboring low- (n = 30) and high-grade gliomas (n = 115), non-glial primary brain tumors (n = 19), radiation necrosis (n = 2), miscellaneous (n = 10) and metastases (n = 8). Four machine-learning models were trained to classify tumor type, grade, glioma margins, and IDH mutation. RESULTS: Using random forests and multilayer perceptrons, the classifiers achieve average test accuracies of 84-87%, 96.1%, 86%, and 91% respectively. All five fluorophore abundances vary between tumor margin types and tumor grades (p < 0.01). For tissue type, at least four of the five fluorophore abundances are significantly different (p < 0.01) between all classes. CONCLUSIONS: These results demonstrate the fluorophores' differing abundances in different tissue classes and the value of the five fluorophores as potential optical biomarkers, opening new opportunities for intraoperative classification systems in fluorescence-guided neurosurgery.

Complete surgical removal of some primary brain tumors is difficult because it can be hard to distinguish the edge of the tumor. We evaluated whether the edges of tumors and the tumor type and grade can be more accurately determined if the tumor is imaged using many different wavelengths of light. We used measurements taken from the tumors of people undergoing brain tumor surgery and developed machine-learning algorithms that could predict where the edge of the tumor was. The methods could also provide information about the type and grade of the brain tumor. These classifications could potentially be used during operations to remove brain tumors more accurately and thus improve the outcome of surgery for people with brain tumors.

Protoporphyrin IX in serum of high-grade glioma patients: A novel target for disease monitoring via liquid biopsy.

High-grade gliomas (HGG) carry a dismal prognosis. Diagnosis comprises MRI followed by histopathological evaluation of tissue; no blood biomarker is available. Patients are subjected to serial MRIs and, if unclear, surgery for monitoring of tumor recurrence, which is laborious. MRI provides only limited diagnostic information regarding the differentiation of true tumor progression from therapy-associated side effects. 5-aminolevulinic acid (5-ALA) is routinely used for induction of protoporphyrin IX (PpIX) accumulation in malignant glioma tissue, enabling improved tumor visualization during fluorescence-guided resection (FGR). We investigated whether PpIX can also serve as a serum HGG marker to monitor relapse. Patients (HGG: n = 23 primary, pHGG; n = 5 recurrent, rHGG) undergoing FGR received 5-ALA following standard clinical procedure. The control group of eight healthy volunteers (HCTR) also received 5-ALA. Serum was collected before and repeatedly up to 72 h after drug administration. Significant PpIX accumulation in HGG was observed after 5-ALA administration (ANOVA: p = 0.005, post-hoc: HCTR vs. pHGG p = 0.029, HCTR vs. rHGG p = 0.006). Separation of HCTR from pHGG was possible when maximum serum PpIX levels were reached (CI95% of tMax). ROC analysis of serum PpIX within CI95% of tMax showed successful classification of HCTR and pHGG (AUCROC 0.943, CI95% 0.884-1.000, p < 0.001); the optimal cut-off for diagnosis was 1275 pmol PpIX/ml serum, reaching 87.0% accuracy, 90.5% positive predictive and 84.0% negative predictive value. Baseline PpIX level was similar in patient and control groups. Thus, 5-ALA is required for PpIX induction, which is safe at the standard clinical dosage. PpIX is a new target for liquid biopsy in glioma. More extensive clinical studies are required to characterize its full potential.

Unraveling the blue shift in porphyrin fluorescence in glioma: The 620 nm peak and its potential significance in tumor biology.

In glioma surgery, the low-density infiltration zone of tumors is difficult to detect by any means. While, for instance, 5-aminolevulinic acid (5-ALA)-induced fluorescence is a well-established surgical procedure for maximizing resection of malignant gliomas, a cell density in tumor tissue of 20-30% is needed to observe visual fluorescence. Hyperspectral imaging is a powerful technique for the optical characterization of brain tissue, which accommodates the complex spectral properties of gliomas. Thereby, knowledge about the signal source is essential to generate specific separation (unmixing) procedures for the different spectral characteristics of analytes and estimate compound abundances. It was stated that protoporphyrin IX (PpIX) fluorescence consists mainly of emission peaks at 634 nm (PpIX634) and 620 nm (PpIX620). However, other members of the substance group of porphyrins fluoresce similarly to PpIX due to their common tetrapyrrole core structure. While the PpIX634 signal has reliably been assigned to PpIX, it has not yet been analyzed if PpIX620 might result from a different porphyrin rather than being a second photo state of PpIX. We thus reviewed more than 200,000 spectra from various tumors measured in almost 600 biopsies of 130 patients. Insufficient consideration of autofluorescence led to artificial inflation of the PpIX620 peak in the past. Recently, five basis spectra (PpIX634, PpIX620, flavin, lipofuscin, and NADH) were described and incorporated into the analysis algorithm, which allowed more accurate unmixing of spectral abundances. We used the improved algorithm to investigate the PpIX620 signal more precisely and investigated coproporphyrin III (CpIII) fluorescence phantoms for spectral unmixing. Our findings show that the PpIX634 peak was the primary source of the 5-ALA-induced fluorescence. CpIII had a similar spectral characteristic to PpIX620. The supplementation of 5-ALA may trigger the increased production of porphyrins other than PpIX within the heme biosynthesis pathway, including that of CpIII. It is essential to correctly separate autofluorescence from the main PpIX634 peak to analyze the fluorescence signal. This article highlights the need for a comprehensive understanding of the spectral complexity in gliomas and suggests less significance of the 620 nm fluorescence peak for PpIX analysis and visualization.

Challenges in, and recommendations for, hyperspectral imaging in ex vivo malignant glioma biopsy measurements.

The visualization of protoporphyrin IX (PPIX) fluorescence with the help of surgical microscopes during 5-aminolevulinic acid-mediated fluorescence-guided resection (FGR) of gliomas is still limited at the tumor margins. Hyperspectral imaging (HI) detects PPIX more sensitively but is not yet ready for intraoperative use. We illustrate the current status with three experiments and summarize our own experience using HI: (1) assessment of HI analysis algorithm using pig brain tissue, (2) a partially retrospective evaluation of our experience from HI projects, and (3) device comparison of surgical microscopy and HI. In (1), we address the problem that current algorithms for evaluating HI data are based on calibration with liquid phantoms, which have limitations. Their pH is low compared to glioma tissue; they provide only one PPIX photo state and only PPIX as fluorophore. Testing the HI algorithm with brain homogenates, we found proper correction for optical properties but not pH. Considerably more PPIX was measured at pH 9 than at pH 5. In (2), we indicate pitfalls and guide HI application. In (3), we found HI superior to the microscope for biopsy diagnosis (AUC = 0.845 ± 0.024 (cut-off 0.75 µg PPIX/ml) vs. 0.710 ± 0.035). HI thus offers potential for improved FGR.

Development and validation of a triple-LED surgical loupe device for fluorescence-guided resections with 5-ALA.

OBJECTIVE: Fluorescence-guided resections performed using 5-aminolevulinic acid (5-ALA) have been studied extensively using the BLUE400 system. The authors introduce a triple-light-emitting diode (LED) headlight/loupe device for visualizing fluorescence, and compare this to the BLUE400 gold standard in order to assure similar and not more or less sensitive protoporphyrin-IX visualization. METHODS: The authors defined the spectral requirements for a triple-LED headlight/loupe device for reproducing the xenon-based BLUE400 module. The system consisted of a white LED (normal surgery), a 409-nm LED for excitation, a 450-nm LED for background illumination, and appropriate observation filters. The prototype's excitation and emission spectra, illumination and detection intensities, and spot homogeneity were determined. The authors further performed a prospectively randomized and blinded study for fluorescence assessments of fresh, marginal, fluorescing and nonfluorescing tumor samples comparing the LED/loupe device with BLUE400 in patients with malignant glioma treated with 20 mg/kg body weight 5-ALA. Tumor samples were immediately assessed in turn, both with a Kinevo and with a novel triple-LED/loupe device by different surgeons. RESULTS: Seven triple-LED/loupe devices were analyzed. Illumination intensities in the 409- and 450-nm range were comparable to BLUE400, with high spot homogeneity. Fluorescence intensities measured distally to microscope oculars/loupes were 9.9-fold higher with the loupe device. For validation 26 patients with malignant gliomas with 240 biopsies were analyzed. With BLUE400 results as the reference, sensitivity for reproducing fluorescence findings was 100%, specificity was 95%, positive predictive value was 98%, negative predictive value was 100%, and accuracy was 95%. This study reached its primary aim, with agreement in 226 of 240 (94.2%, 95% CI 0.904-0.968). CONCLUSIONS: The authors observed only minor differences regarding spectra and illumination intensities during evaluation. Fluorescence intensities available to surgeons were 9.9-fold higher with the loupe device. Importantly, the independent perception of fluorescence achieved using the new system and BLUE400 was statistically equivalent. The authors believe the triple-LED/loupe device to be a useful and safe option for surgeons who prefer loupes to the microscope for resections in appropriate patients.

Characterization of autofluorescence and quantitative protoporphyrin IX biomarkers for optical spectroscopy-guided glioma surgery.

5-Aminolevulinic acid (5-ALA)-mediated fluorescence does not effectively depict low grade gliomas (LGG) or the infiltrative tumor portion of high-grade gliomas (HGG). While spectroscopy improves sensitivity and precision, this is currently limited by autofluorescence and a second protoporphyrin IX (PpIX) fluorescence state at 620 nm. We investigated the autofluorescence to better characterize the present spectra and thus increase PpIX quantification precision and sensitivity. This study included 128 patients undergoing surgery for malignant glioma. 5-ALA (Gliolan) was administered before anesthesia, and fluorescence was measured using a hyperspectral device. It was found that all 2692 measured spectra consisted of contributions from 620 to 634 nm PpIX, NADH, lipofuscin, and flavins. The basis spectra were characterized and their use in spectral unmixing led to 82.4% lower fitting error for weakly fluorescing areas (p < 0.001), and 92.3% fewer false positive tumor identifications in control measurements (p = 0.0065) compared to previous works. They also decreased the PpIX620 contribution, thus halving the mean Ratio620/634 (p < 0.001). The ratio was approximately 0 for HGGs and increasing for LGGs, as demonstrated previously. Additionally, the Ratio620/634, the MIB-1/Ki-67 proliferation index, and the PpIX peak blue-shift were found to be significantly related to WHO grade, fluorescence visibility, and PpIX contribution (p < 0.001), and the value of these three as quantitative biomarkers is discussed.