Optical coherence tomography (OCT) shows potential as an intraoperative guidance tool. However, OCT images are difficult to interpret and real-time analysis methods are needed to promote its clinical use. This study investigates deep learning-based OCT image classification with application on thyroid diseases. To evaluate the impact of data pre-processing and model architecture on classification performance, 2D and 3D deep learning models were implemented and trained on OCT data from ex-vivo thyroid samples. For 2D classification, deeper models and the ones using information from different spatial resolution achieved highest performance. However, 3D models outperform the 2D counterparts in most classification tasks.
Optical coherence tomography (OCT) can provide exquisite details of tissue microstructure without traditional tissue sectioning, with potential diagnostic and intraoperative applications in a variety of clinical areas. In thyroid surgery, OCT could provide information to reduce the risk of damaging normal tissue. Thyroid tissue's follicular structure alters in case of various pathologies including the non-malignant ones which can be imaged using OCT. The success of deep learning for medical image analysis encourages its application on OCT thyroid images for quantitative analysis of tissue microstructure. To investigate the potential of a deep learning approach to segment the follicular structure in OCT images, a 2D U-Net was trained on b-scan OCT images acquired from ex vivo adult human thyroid samples a effected by a range of pathologies. Results on a pool of 104 annotated images showed a mean Dice score of 0.74±0.19 and 0.92±0.09 when segmenting the follicular structure and the surrounding tissue on the test dataset (n=10 images). This study shows that a deep learning approach for tissue microstructure segmentation in OCT images is possible. The achieved performance without requiring manual intervention encourages the application of a deep-learning method for real-time analysis of OCT data.
In the routine of stereotactic biopsy on suspected tumors located deep in the brain or patients with multiple lesions, tissue samples are harvested to determine the type of malignancy. Biopsies are taken from pre-calculated positions based on the preoperative radiologic images susceptible to brain shift. In such cases the biopsy procedure may need to be repeated leading to a longer operation time. To provide guidance for targeting diagnostic tumor tissue and to avoid vessel rupture on the insertion path of the tumor, an application specific fiber optic probe was developed. The setup incorporated spectroscopy for 5-aminolevulinic acid induced protopophyrin IX (PpIX) fluorescence in the tumor and laser Doppler for measuring microvascular blood flow which recorded backscattered light (TLI) at 780 nm and blood perfusion. The recorded signals were compared to the histopathologic diagnosis of the tissue samples (n=16) and to the preoperative radiologic images. All together 146 fluorescence and 276 laser Doppler signals were recorded along 5 trajectories in 4 patients. On all occasions strong PpIX fluorescence peaks were visible during real-time guidance. Comparing the gliotic tumor marginal zone with the tumor, the PpIX (51 vs. 528 a.u., [0-1790], p < 0.05) was higher and TLI (2.9 vs. 2.0 a.u., [0-4.1], p < 0.05) was lower in tumor. The autofluorescence (104 vs.70 a.u., [0-442], p > 0.05) and blood perfusion (8.3 vs. 17 a.u., [0-254], p > 0.05) were not significantly different. In conclusion, the optical guidance probe made real-time tumor detection and vessel tracking possible during the stereotactic biopsy procedures. Moreover, the fluorescence and blood perfusion in the tumor could be studied at controlled positions in the brain and the tumor.
5-aminolevulinic-acid-(5-ALA)-induced protoporphyrin IX (PpIX) fluorescence may be used to improve stereotactic brain tumor biopsies. In this study, the sensitivity of PpIX-based tumor detection has been investigated for two potential excitation wavelengths (405 nm, 633 nm). Using a 200 μm fiber in contact with semi-infinite optical phantoms containing ink and Lipovenös, PpIX detection limits of 4.0 nM and 200 nM (relating to 1 mW excitation power) were determined for 405 nm and 633 nm excitation, respectively. Hence, typical PpIX concentrations in glioblastomas of a few μM should be well detectable with both wavelengths. Additionally, blood layers of selected thicknesses were placed between fiber and phantom. Red excitation was shown to be considerably less affected by blood interference: A 50 μm blood layer, for instance, blocked the 405- nm-excited fluorescence completely, but reduced the 633-nm-excited signal by less than 50%. Ray tracing simulations demonstrated that – without blood layer – the sensitivity advantage of 405 nm rises for decreasing fluorescent volume from 50-fold to a maximum of 100-fold. However, at a tumor volume of 1 mm3, which is a typical biopsy sample size, the 633-nm-excited fluorescence signal is only reduced by about 10%. Further simulations revealed that with increasing fiber-tumor distance, the signal drops faster for 405 nm. This reduces the risk of detecting tumor tissue outside the needle’s coverage, but diminishes the overlap between optically and mechanically sampled volumes. While 405 nm generally offers a higher sensitivity, 633 nm is more sensitive to distant tumors and considerably superior in case of blood-covered tumor tissue.
Stereotactic biopsy procedure is performed to obtain a tissue sample for diagnosis purposes. Currently, a fiber-based mechano-optical device for stereotactic biopsies of brain tumors is developed. Two different fluorophores are employed to improve the safety and reliability of this procedure: The fluorescence of intravenously applied indocyanine green (ICG) facilitates the recognition of blood vessels and thus helps minimize the risk of cerebral hemorrhages. 5- aminolevulinic-acid-induced protoporphyrin IX (PpIX) fluorescence is used to localize vital tumor tissue. ICG fluorescence detection using a 2-fiber probe turned out to be an applicable method to recognize blood vessels about 1.5 mm ahead of the fiber tip during a brain tumor biopsy. Moreover, the suitability of two different PpIX excitation wavelengths regarding practical aspects was investigated: While PpIX excitation in the violet region (at 405 nm) allows for higher sensitivity, red excitation (at 633 nm) is noticeably superior with regard to blood layers obscuring the fluorescence signal. Contact measurements on brain simulating agar phantoms demonstrated that a typical blood coverage of the tumor reduces the PpIX signal to about 75% and nearly 0% for 633 nm and 405 nm excitation, respectively. As a result, 633 nm seems to be the wavelength of choice for PpIX-assisted detection of high-grade gliomas in stereotactic biopsy.
Fluorescence guidance using 5-aminolevulinic acid (5-ALA) for brain tumor resection is a recent technique applied to the highly malignant brain tumors. Five-ALA accumulates as protoporphyrin IX fluorophore in the tumor cells in different concentrations depending on the tumor environment and cell properties. Our group has developed a fluorescence spectroscopy system used with a hand-held probe intra-operatively. The system has shown improvement of fluorescence detection and allows quantification that preliminarily correlates with tumor malignancy grade during surgery. However, quantification of fluorescence is affected by several factors including the initial fluorophore concentration, photobleaching due to operating lamps and attenuation from the blood. Accordingly, an optical phantom was developed to enable controlled fluorescence measurements and evaluation of the system outside of the surgical procedure. The phantom mimicked the optical properties of glioma at the specific fluorescence excitation wavelength when different concentrations of the fluorophore were included in the phantom. To allow evaluation of photobleaching, kinetics of fluorophore molecules in the phantom was restricted by solidifying the phantoms. Moreover, a model for tissue autofluorescence was added. The fluorescence intensity’s correlation with fluorophore concentration in addition to the photobleaching properties were investigated in the phantoms and were compared to the clinical data measured on the brain tumor.
The principles of cancer treatment has for years been radical resection of the primary tumor. In the oncologic surgeries where the affected cancer site is close to the lymphatic system, it is as important to detect the draining lymph nodes for metastasis (lymph node mapping). As a replacement for conventional radioactive labeling, indocyanine green (ICG) has shown successful results in lymph node mapping; however, most of the ICG fluorescence detection techniques developed are based on camera imaging. In this work, fluorescence spectroscopy using a fiber-optical probe was evaluated on a tissue-like ICG phantom with ICG concentrations of 6-64 μM and on breast tissue from five patients. Fiber-optical based spectroscopy was able to detect ICG fluorescence at low intensities; therefore, it is expected to increase the detection threshold of the conventional imaging systems when used intraoperatively. The probe allows spectral characterization of the fluorescence and navigation in the tissue as opposed to camera imaging which is limited to the view on the surface of the tissue.
Fluorescence guidance in brain tumor resection is performed intra-operatively where bleeding is included. When using
fiber-optical probes, the transmission of light to and from the tissue is totally or partially blocked if a small amount of
blood appears in front of the probe. Sometimes even after rinsing with saline, the remnant blood cells on the optical
probe head, disturb the measurements. In such a case, the corresponding spectrum cannot be reliably quantified and is
therefore discarded. The optimal case would be to calculate and take out the blood effect systematically from the
collected signals. However, the first step is to study the pattern of blood interference in the fluorescence spectrum. In this
study, a fiber-optical based fluorescence spectroscopy system with a laser excitation light of 405 nm (1.4 J/cm2) was
used during fluorescence guided brain tumor resection using 5-aminolevulinic acid (5-ALA). The blood interference
pattern in the fluorescence spectrum collected from the brain was studied in two patients. The operation situation was
modeled in the laboratory by placing blood drops from the finger tip on the skin of forearm and the data was compared to
the brain in vivo measurements. Additionally, a theoretical model was developed to simulate the blood interference
pattern on the skin autofluorescence. The blood affects the collected fluorescence intensity and leaves traces of oxy and
deoxy-hemoglobin absorption peaks. According to the developed theoretical model, the autofluorescence signal is
considered to be totally blocked by an approximately 500 μm thick blood layer.
The highly malignant brain tumor, glioblastoma multiforme (GBM), is difficult to fully delineate during surgical
resection due to its infiltrative ingrowth and morphological similarities to surrounding functioning brain tissue.
Selectiveness of GBM to 5-aminolevulinic acid (5-ALA) induced protoporphyrin IX (PpIX) is reported by other
researchers to visualize tumor margins under blue light microscopy. To allow objective detection of GBM, a compact
and portable fiber optic based fluorescence spectroscopy system is developed. This system is able to deliver excitation
laser light (405 nm) in both the continuous and pulsed mode. PpIX fluorescence peaks are detected at 635 and 704 nm,
using a fiber-coupled spectrometer. It is necessary to optimize the detection efficiency of the system as the PpIX quickly
photobleaches during the laser illumination. A light dose of 2.5 mJ (fluence rate = 9 mJ/mm2) is experimentally approved
to excite an acceptable level of fluourescence signal arising from glioblastoma. In pulsed illumination mode, an
excitation dose of 2.5 mJ, with a dark interval of 0.5 s (duty cycle 50%) shows a significantly shorter photobleaching
time in comparison to the continuous illumination mode with the same laser power (p < 0.05). To avoid photobleaching
(the remaining signal is more than 90% of its initial value) when measuring with 2.5 mJ delivered energy, the time for
continuous and pulsed illumination should be restricted to 2.5 and 1.1 s, respectively.
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.