Nicolai Oetter

Automatic Classification of Cancerous Tissue in Laserendomicroscopy Images of the Oral Cavity using Deep Learning

Oral Squamous Cell Carcinoma (OSCC) is a common type of cancer of the oral epithelium. Despite their high impact on mortality, sufficient screening methods for early diagnosis of OSCC often lack accuracy and thus OSCCs are mostly diagnosed at a late stage. Early detection and accurate outline estimation of OSCCs would lead to a better curative outcome and a reduction in recurrence rates after surgical treatment. Confocal Laser Endomicroscopy (CLE) records sub-surface micro-anatomical images for in vivo cell structure analysis. Recent CLE studies showed great prospects for a reliable, real-time ultrastructural imaging of OSCC in situ. We present and evaluate a novel automatic approach for OSCC diagnosis using deep learning technologies on CLE images. The method is compared against textural feature-based machine learning approaches that represent the current state of the art. For this work, CLE image sequences (7894 images) from patients diagnosed with OSCC were obtained from 4 specific locations in the oral cavity, including the OSCC lesion. The present approach is found to outperform the state of the art in CLE image recognition with an area under the curve (AUC) of 0.96 and a mean accuracy of 88.3% (sensitivity 86.6%, specificity 90%).

Tissue Discrimination by Uncorrected Autofluorescence Spectra: A Proof-of-Principle Study for Tissue-Specific Laser Surgery

Laser surgery provides a number of advantages over conventional surgery. However, it implies large risks for sensitive tissue structures due to its characteristic non-tissue-specific ablation. The present study investigates the discrimination of nine different ex vivo tissue types by using uncorrected (raw) autofluorescence spectra for the development of a remote feedback control system for tissue-selective laser surgery. Autofluorescence spectra (excitation wavelength 377 ± 50 nm) were measured from nine different ex vivo tissue types, obtained from 15 domestic pig cadavers. For data analysis, a wavelength range between 450 nm and 650 nm was investigated. Principal Component Analysis (PCA) and Quadratic Discriminant Analysis (QDA) were used to discriminate the tissue types. ROC analysis showed that PCA, followed by QDA, could differentiate all investigated tissue types with AUC results between 1.00 and 0.97. Sensitivity reached values between 93% and 100% and specificity values between 94% and 100%. This ex vivo study shows a high differentiation potential for physiological tissue types when performing autofluorescence spectroscopy followed by PCA and QDA. The uncorrected autofluorescence spectra are suitable for reliable tissue discrimination and have a high potential to meet the challenges necessary for an optical feedback system for tissue-specific laser surgery.

Classification of Confocal Laser Endomicroscopic Images of the Oral Cavity to Distinguish Pathological from Healthy Tissue

Confocal laser endomicroscopy is a recently introduced advanced imaging technique which enables microscopic imaging of the mucosa in-vivo. This technique has already been applied successfully during diagnosis of gastrointestinal diseases. Whereas for this purpose several computer aided diagnosis approaches exist, we present a classification system that is able to differentiate between healthy and pathological images of the oral cavity. Varying textural features of small rectangular regions are evaluated using random forests and support vector machines. Preliminary results reach up to 99.2% classification rate. This indicates that an automatic classification system to differentiate between healthy and pathological mucosa of the oral cavity is feasible.

Gingival esthetics and oral health-related quality of life in patients with cleft lip and palate

While the oral health-related quality of life (OHRQoL) is known to be reduced in patients with cleft lip and palate (CLP), its inter-dependency with the soft tissue characteristics of the CLP area remains unclear. This study aimed to evaluate the soft tissue characteristics in the treated cleft area in order to investigate whether gingival esthetics correlate with OHRQoL. Thirty-six patients with unilateral or bilateral CLP (46 cleft areas) were investigated after secondary/tertiary alveolar bone grafting and orthodontic/prosthetic implant treatment using an adapted score to rate gingival esthetics (clinical esthetic score, CES). The patients OHRQoL was determined using the German short version of the Oral Health Impact Profile questionnaire (OHIP-G14). The results showed a significantly better rating in patients with their own teeth in situ (12.05±1.10) than in patients with implants (6.95±4.78) or prosthetics (4.00±3.58). The best OHRQoL values were achieved by patients with their own teeth integrated into the cleft area (1.32±2.31), followed by patients with implants (2.33±2.33) and prosthetics (3.75±5.87). A significant (P=0.017) correlation was found between OHIP-G14 and CES scores, suggesting an increased OHRQoL in cases with higher oral esthetics in the cleft area. The therapeutic strategy contributes to both gingival esthetics and OHRQoL. The patients subjective perception of OHRQoL can be attributed to objective gingival esthetic ratings.

Investigation of Laser Induced Breakdown Spectroscopy (LIBS) for the Differentiation of Nerve and Gland Tissue—A Possible Application for a Laser Surgery Feedback Control Mechanism

Laser surgery provides clean, fast and accurate modeling of tissue. However, the inability to determine what kind of tissue is being ablated at the bottom of the cut may lead to the iatrogenic damage of structures that were meant to be preserved. In this context, nerve preservation is one of the key challenges in any surgical procedure. One example is the treatment of parotid gland pathologies, where the facial nerve (N. VII) and its main branches run through and fan out inside the glands parenchyma. A feedback system that automatically stops the ablation to prevent nerve-tissue damage could greatly increase the applicability and safety of surgical laser systems. In the present study, Laser Induced Breakdown Spectroscopy (LIBS) is used to differentiate between nerve and gland tissue of an ex-vivo pig animal model. The LIBS results obtained in this preliminary experiment suggest that the measured spectra, containing atomic and molecular emissions, can be used to differentiate between the two tissue types. The measurements and differentiation were performed in open air and under normal stray light conditions.

Motion Artifact Detection in Confocal Laser Endomicroscopy Images

Confocal Laser Endomicroscopy (CLE), an optical imaging technique allowing non-invasive examination of the mucosa on a (sub)cellular level, has proven to be a valuable diagnostic tool in gastroenterology and shows promising results in various anatomical regions including the oral cavity. Recently, the feasibility of automatic carcinoma detection for CLE images of sufficient quality was shown. However, in real world data sets a high amount of CLE images is corrupted by artifacts. Amongst the most prevalent artifact types are motion-induced image deteriorations. In the scope of this work, algorithmic approaches for the automatic detection of motion artifact-tainted image regions were developed. Hence, this work provides an important step towards clinical applicability of automatic carcinoma detection. Both, conventional machine learning and novel, deep learning-based approaches were assessed. The deep learning-based approach outperforms the conventional approaches, attaining an AUC of 0.90.

Patch-based Carcinoma Detection on Confocal Laser Endomicroscopy Images - A Cross-site Robustness Assessment

Deep learning technologies such as convolutional neural networks (CNN) provide powerful methods for image recognition and have recently been employed in the field of automated carcinoma detection in confocal laser endomicroscopy (CLE) images. CLE is a (sub-)surface microscopic imaging technique that reaches magnifications of up to 1000x and is thus suitable for in vivo structural tissue analysis. In this work, we aim to evaluate the prospects of a priorly developed deep learning-based algorithm targeted at the identification of oral squamous cell carcinoma with regard to its generalization to further anatomic locations of squamous cell carcinomas in the area of head and neck. We applied the algorithm on images acquired from the vocal fold area of five patients with histologically verified squamous cell carcinoma and presumably healthy control images of the clinically normal contra-lateral vocal cord. We find that the network trained on the oral cavity data reaches an accuracy of 89.45% and an area-under-the-curve (AUC) value of 0.955, when applied on the vocal cords data. Compared to the state of the art, we achieve very similar results, yet with an algorithm that was trained on a completely disjunct data set. Concatenating both data sets yielded further improvements in cross-validation with an accuracy of 90.81% and AUC of 0.970. In this study, for the first time to our knowledge, a deep learning mechanism for the identification of oral carcinomas using CLE Images could be applied to other disciplines in the area of head and neck. This study shows the prospect of the algorithmic approach to generalize well on other malignant entities of the head and neck, regardless of the anatomical location and furthermore in an examiner-independent manner.

Deep learning-based detection of motion artifacts in probe-based confocal laser endomicroscopy images

Purpose:Probe-based confocal laser endomicroscopy (pCLE) is a subcellular in vivo imaging technique capable of producing images that enable diagnosis of malign structural modifications in epithelial tissue. Images acquired with pCLE are, however, often tainted by significant artifacts that impair diagnosis. This is especially detrimental for automated image analysis, which is why said images are often excluded from recognition pipelines.MethodsWe present an approach for the automatic detection of motion artifacts in pCLE images and apply this methodology to a data set of 15 thousand images of epithelial tissue acquired in the oral cavity and the vocal folds. The approach is based on transfer learning from intermediate endpoints within a pre-trained Inception v3 network with tailored preprocessing. For detection within the non-rectangular pCLE images, we perform pooling within the activation maps of the network and evaluate this at different network depths.ResultsWe achieved area under the ROC curve values of 0.92 with the proposed method, compared to 0.80 for the best feature-based machine learning approach. Our overall accuracy with the presented approach is 94.8%.ConclusionOver traditional machine learning approaches with state-of-the-art features, we achieved significantly improved overall performance.

Autofluorescence spectroscopy for nerve-sparing laser surgery of the head and neck—the influence of laser-tissue interaction

The use of remote optical feedback systems represents a promising approach for minimally invasive, nerve-sparing laser surgery. Autofluorescence properties can be exploited for a fast, robust identification of nervous tissue. With regard to the crucial step towards clinical application, the impact of laser ablation on optical properties in the vicinity of structures of the head and neck has not been investigated up to now. We acquired 24,298 autofluorescence spectra from 135 tissue samples (nine ex vivo tissue types from 15 bisected pig heads) both before and after ER:YAG laser ablation. Sensitivities, specificities, and area under curve(AUC) values for each tissue pair as well as the confusion matrix were statistically calculated for pre-ablation and post-ablation autofluorescence spectra using principal component analysis (PCA), quadratic discriminant analysis (QDA), and receiver operating characteristics (ROC). The confusion matrix indicated a highly successful tissue discrimination rate before laser exposure, with an average classification error of 5.2%. The clinically relevant tissue pairs nerve/cancellous bone and nerve/salivary gland yielded an AUC of 100% each. After laser ablation, tissue discrimination was feasible with an average classification accuracy of 92.1% (average classification error 7.9%). The identification of nerve versus cancellous bone and salivary gland performed very well with an AUC of 100 and 99%, respectively. Nerve-sparing laser surgery in the area of the head and neck by means of an autofluorescence-based feedback system is feasible even after ER-YAG laser-tissue interactions. These results represent a crucial step for the development of a clinically applicable feedback tool for laser surgery interventions in the oral and maxillofacial region.

Long-term Outcome of Speech Intelligibility in Maxillary Dental Rehabilitation with Full Dentures: A Prospective Study Using Automatic Speech Quantification

AIMS Complete maxillary edentulism and prosthetic rehabilitation with removable full dentures are known to affect speech intelligibility. The aim of this study was to prospectively investigate the long-term effect of time on speech intelligibility in patients being rehabilitated with newly fabricated full maxillary dentures. MATERIALS AND METHODS Speech was recorded in a group of 14 patients (male = 9, female = 5; mean age ± standard deviation [SD] = 66.14 ± 7.03 years) five times within a mean period of 4 years (mean ± SD: 47.50 ± 18.16 months; minimum/maximum: 24/68 months) and in a control group of 40 persons with healthy dentition (male = 30, female = 10; mean age ± SD = 59 ± 12 years). All 14 participants had their inadequate removable full maxillary dentures replaced with newly fabricated dentures. Speech intelligibility was measured by means of a polyphone-based speech recognition system that automatically computed the percentage of accurately spoken words (word accuracy [WA]) at five different points in time: 1 week prior to prosthetic maxillary rehabilitation (both with and without inadequate dentures in situ) and at 1 week, 6 months, and a mean of 48 months after the insertion of newly fabricated full maxillary dentures. RESULTS Speech intelligibility of the patients significantly improved after 6 months of adaptation to the new removable full maxillary dentures (WA = 66.93% ± 9.21%) compared to inadequate dentures in situ (WA = 60.12% ± 10.48%). After this period, no further significant change in speech intelligibility was observed. After 1 week of adaptation, speech intelligibility of the rehabilitated patients aligned with that of the control group (WA = 69.79% ± 10.60%) and remained at this level during the examination period of 48 months. CONCLUSION The provision of new removable full maxillary dentures can improve speech intelligibility to the level of a healthy control group on a long-term basis.