Comparative Impact of Various Fasting Periods on the Welfare of Sprague-Dawley Rats With or Without Supplementation.

In nonclinical toxicology studies, lab animals are fasted typically overnight, to reduce variability in some clinical pathology parameters. However, fasting adds undue stress, and this is particularly concerning in rodents given their fast metabolic rates. Furthermore, as rodents are nocturnal animals, an overnight fasting may cause a protracted negative metabolic state even when the fasting has technically ended, given their minimal activity and food consumption during the day. Therefore, to evaluate the impacts of different fasting durations (±DietGel supplementation) on rats' welfare, we assessed the traditional and ancillary clinical pathology parameters in Sprague-Dawley rats, along with body weight, organ weight, and histopathology. Although most endpoints were comparable between the different fasting durations (±DietGel supplementation), the long fasting times (≥8 hr) without DietGel supplementation caused significant decreases in body weight, liver weight, liver glycogen content, serum glucose, triglyceride, and creatinine concentrations-all findings suggestive of a negative energy balance that could impact animal welfare and consequently, data quality; while the short fasting time (4 hr) and DietGel supplementation were associated with higher triglycerides variability. Hence, we propose that short fasting time should be adequate for most toxicology studies in rats, and long fasting times should only be accommodated with scientific justification.

Toxicologic Pathology Forum: A Roadmap for Building State-of-the-Art Digital Image Data Resources for Toxicologic Pathology in the Pharmaceutical Industry.

Digitization of histologic slides brings with it the promise of enhanced toxicologic pathology practice through the increased application of computational methods. However, the development of these advanced methods requires access to substrate image data, that is, whole slide images (WSIs). Deep learning methods, in particular, rely on extensive training data to develop robust algorithms. As a result, pharmaceutical companies interested in leveraging computational methods in their digital pathology workflows must first invest in data infrastructure to enable data access for both data scientists and pathologists. The process of building robust image data resources is challenging and includes considerations of generation, curation, and storage of WSI files, and WSI access including via linked metadata. This opinion piece describes the collective experience of building resources for WSI data in the Roche group. We elaborate on the challenges encountered and solutions developed with the goal of providing examples of how to build a data resource for digital pathology analytics in the pharmaceutical industry.

Impact of color augmentation and tissue type in deep learning for hematoxylin and eosin image super resolution.

Single image super-resolution is an important computer vision task with applications including remote sensing, medical imaging, and surveillance. Modern work on super-resolution utilizes deep learning to synthesize high resolution (HR) images from low resolution images (LR). With the increased utilization of digitized whole slide images (WSI) in pathology workflows, digital pathology has emerged as a promising domain for super-resolution. Despite extensive existing research into super-resolution, there remain challenges specific to digital pathology. Here, we investigated image augmentation techniques for hematoxylin and eosin (H&E) WSI super-resolution and model generalizability across diverse tissue types. In addition, we investigated shortcomings with common quality metrics (peak signal-to-noise ratio (PSNR), structure similarity index (SSIM)) by conducting a perceptual quality survey for super-resolved pathology images. High performing deep super-resolution models were used to generate 20X HR images from LR images (5X or 10X equivalent) for 11 different tissues and 30 human evaluators were asked to score the quality of the generated versus the ground truth 20X HR images. The scores given by a human rater and the PSNR or the SSIM were compared to investigate the correlation between model training parameters. We found that models trained on multiple tissues generalized better than those trained on a single tissue type. We also found that PSNR correlated with perceptual quality (R = 0.26) less accurately than did SSIM (R = 0.64), suggesting that the SSIM quality metric is insufficient. The methods proposed in this study can be used to virtually magnify H&E images with better perceptual quality than interpolation methods (i.e., bicubic interpolation) commonly implemented in digital pathology software. The impact of deep SISR methods is more notable when scaling to 4X is needed, such as in the case of super-resolving a low magnification WSI from 10X to 40X.

Multiscale generative model using regularized skip-connections and perceptual loss for anomaly detection in toxicologic histopathology.

Background: Automated anomaly detection is an important tool that has been developed for many real-world applications, including security systems, industrial inspection, and medical diagnostics. Despite extensive use of machine learning for anomaly detection in these varied contexts, it is challenging to generalize and apply these methods to complex tasks such as toxicologic histopathology (TOXPATH) assessment (i.e.,finding abnormalities in organ tissues). In this work, we introduce an anomaly detection method using deep learning that greatly improves model generalizability to TOXPATH data. Methods: We evaluated a one-class classification approach that leverages novel regularization and perceptual techniques within generative adversarial network (GAN) and autoencoder architectures to accurately detect anomalous histopathological findings of varying degrees of complexity. We also utilized multiscale contextual data and conducted a thorough ablation study to demonstrate the efficacy of our method. We trained our models on data from normal whole slide images (WSIs) of rat liver sections and validated on WSIs from three anomalous classes. Anomaly scores are collated into heatmaps to localize anomalies within WSIs and provide human-interpretable results. Results: Our method achieves 0.953 area under the receiver operating characteristic on a real-worldTOXPATH dataset. The model also shows good performance at detecting a wide variety of anomalies demonstrating our method's ability to generalize to TOXPATH data. Conclusion: Anomalies in both TOXPATH histological and non-histological datasets were accurately identified with our method, which was only trained with normal data.

Assessment of Skin Toxicity in an in Vitro Reconstituted Human Epidermis Model Using Deep Learning.

Skin toxicity is a common safety concern associated with drugs that inhibit epidermal growth factor receptors as well as other targets involved in epidermal growth and differentiation. Recently, the use of a three-dimensional reconstructed human epidermis model enabled large-scale drug screening and showed potential for predicting skin toxicity. Although a decrease in epidermal thickness was often observed when the three-dimensional reconstructed tissues were exposed to drugs causing skin toxicity, the thickness evaluation of epidermal layers from a pathologist was subjective and not easily reproducible or scalable. In addition, the subtle differences in thickness among tissues, as well as the large number of samples tested, made cross-study comparison difficult when a manual evaluation strategy was used. The current study used deep learning and image-processing algorithms to measure the viable epidermal thickness from multiple studies and found that the measured thickness was not only significantly correlated with a pathologist's semi-quantitative evaluation but was also in close agreement with the quantitative measurement performed by pathologists. Moreover, a sensitivity of 0.8 and a specificity of 0.75 were achieved when predicting the toxicity of 18 compounds with clinical observations with these epidermal thickness algorithms. This approach is fully automated, reproducible, and highly scalable. It not only shows reasonable accuracy in predicting skin toxicity but also enables cross-study comparison and high-throughput compound screening.

Ovarian Toxicity Assessment in Histopathological Images Using Deep Learning.

As ovarian toxicity is often a safety concern for cancer therapeutics, identification of ovarian pathology is important in early stages of preclinical drug development, particularly when the intended patient population include women of child-bearing potential. Microscopic evaluation by pathologists of hematoxylin and eosin (H&E)-stained tissues is the current gold standard for the assessment of organs in toxicity studies. However, digital pathology and advanced image analysis are being explored with greater frequency and broader applicability to tissue evaluations in toxicologic pathology. Our objective in this work was to develop an automated method that rapidly enumerates rat ovarian corpora lutea on standard H&E-stained slides with comparable accuracy to the gold standard assessment by a pathologist. Herein, we describe an algorithm generated by a deep learning network and tested on 5 rat toxicity studies, which included studies that both had and had not previously been diagnosed with effects on number of ovarian corpora lutea. Our algorithm could not only enumerate corpora lutea accurately in all studies but also revealed distinct trends for studies with and without reproductive toxicity. Our method could be a widely applied tool to aid analysis in general toxicity studies.

Distinct Angiogenic Changes during Carcinogenesis Defined by Novel Label-Free Dark-Field Imaging in a Hamster Cheek Pouch Model.

There remain gaps in knowledge concerning how vascular morphology evolves during carcinogenesis. In this study, we imaged neovascularization by label-free dark-field microscopy of a 7,12-Dimethylbenz[a]anthracene (DMBA)-induced hamster cheek pouch model of oral squamous cell carcinoma (SCC). Wavelength-dependent imaging revealed distinct vascular features at different imaging depths and vessel sizes. Vascular tortuosity increased significantly in high-risk lesions, whereas diameter decreased significantly in hyperplastic and SCC lesions. Large vessels preserved the same trends seen in the original images, whereas small vessels displayed different trends, with length and diameter increasing during carcinogenesis. On the basis of these data, we developed and validated a classification algorithm incorporating vascular features from different vessel masks. Receiver operator curves generated from the classification results demonstrated high accuracies in discriminating normal and hyperplasia from high-grade lesions (AUC > 0.94). Overall, these results provided automated imaging of vasculature in the earliest stages of carcinogenesis from which one can extract robust endpoints. The optical toolbox described here is simple, low-cost and portable, and can be used in a variety of health care and research settings for cancer prevention and pharmacology research. Cancer Res; 77(24); 7109-19. ©2017 AACR.

Development of enhanced ethanol ablation as an alternative to surgery in treatment of superficial solid tumors.

While surgery is at the foundation of cancer treatment, its access is limited in low-income countries. Here, we describe development of a low-cost alternative therapy based on intratumoral ethanol injection suitable for resource-limited settings. Although ethanol-based tumor ablation is successful in treating hepatocellular carcinomas, the necessity for multiple treatments, injection of large fluid volumes, and decreased efficacy in treatment of non-capsulated tumors limit its applicability. To address these limitations, we investigated an enhanced ethanol ablation strategy to retain ethanol within the tumor through the addition of ethyl cellulose. This increases the viscosity of injected ethanol and forms an ethanol-based gel-phase upon exposure to the aqueous tumor environment. This technique was first optimized to maximize distribution volume, using tissue-simulating phantoms. Then, chemically-induced epithelial tumors in the hamster cheek pouch were treated. As controls, pure ethanol injections of either four times or one-fourth the tumor volume induced complete regression of 33% and 0% of tumors, respectively. In contrast, ethyl cellulose-ethanol injections of one-fourth the tumor volume induced complete regression in 100% of tumors. These results contribute to proof-of-concept for enhanced ethanol ablation as a novel and effective alternative to surgery for tumor treatment, with relevance to resource-limited settings.

Dark field optical imaging reveals vascular changes in an inducible hamster cheek pouch model during carcinogenesis.

In this study, we propose a low-cost cross-polarized dark field microscopy system for in vivo vascular imaging to detect head and neck cancer. A simple-to-use Gabor-filter-based image processing technique was developed to objectively and automatically quantify several important vascular features, including tortuosity, length, diameter and area fraction, from vascular images. Simulations were performed to evaluate the accuracies of vessel segmentation and feature extraction for our algorithm. Sensitivity and specificity for vessel segmentation of the Gabor masks both remained above 80% at all contrast levels when compared to gold-standard masks. Errors for vascular feature extraction were under 5%. Moreover, vascular contrast and vessel diameter were identified to be the two primary factors which affected the segmentation accuracies. After our algorithm was validated, we monitored the blood vessels in an inducible hamster cheek pouch carcinogen model over 17 weeks and quantified vascular features during carcinogenesis. A significant increase in vascular tortuosity and a significant decrease in vessel length were observed during carcinogenesis.

Oxygen and Perfusion Kinetics in Response to Fractionated Radiation Therapy in FaDu Head and Neck Cancer Xenografts Are Related to Treatment Outcome.

PURPOSE: To test whether oxygenation kinetics correlate with the likelihood for local tumor control after fractionated radiation therapy. METHODS AND MATERIALS: We used diffuse reflectance spectroscopy to noninvasively measure tumor vascular oxygenation and total hemoglobin concentration associated with radiation therapy of 5 daily fractions (7.5, 9, or 13.5 Gy/d) in FaDu xenografts. Spectroscopy measurements were obtained immediately before each daily radiation fraction and during the week after radiation therapy. Oxygen saturation and total hemoglobin concentration were computed using an inverse Monte Carlo model. RESULTS: First, oxygenation kinetics during and after radiation therapy, but before tumor volumes changed, were associated with local tumor control. Locally controlled tumors exhibited significantly faster increases in oxygenation after radiation therapy (days 12-15) compared with tumors that recurred locally. Second, within the group of tumors that recurred, faster increases in oxygenation during radiation therapy (day 3-5 interval) were correlated with earlier recurrence times. An area of 0.74 under the receiver operating characteristic curve was achieved when classifying the local control tumors from all irradiated tumors using the oxygen kinetics with a logistic regression model. Third, the rate of increase in oxygenation was radiation dose dependent. Radiation doses ≤9.5 Gy/d did not initiate an increase in oxygenation, whereas 13.5 Gy/d triggered significant increases in oxygenation during and after radiation therapy. CONCLUSIONS: Additional confirmation is required in other tumor models, but these results suggest that monitoring tumor oxygenation kinetics could aid in the prediction of local tumor control after radiation therapy.

Assessment of the sensitivity and specificity of tissue-specific-based and anatomical-based optical biomarkers for rapid detection of human head and neck squamous cell carcinoma.

OBJECTIVES: We propose the use of morphological optical biomarkers for rapid detection of human head and neck squamous cell carcinoma (HNSCC) by leveraging the underlying tissue characteristics in aerodigestive tracts. MATERIALS AND METHODS: Diffuse reflectance spectra were obtained from malignant and contra-lateral normal tissues of 57 patients undergoing panendoscopy and biopsy. Oxygen saturation, total hemoglobin concentration, and the reduced scattering coefficient were extracted. Differences in malignant and normal tissues were examined based on two different groupings: anatomical site and morphological tissue type. RESULTS AND CONCLUSIONS: Measurements were acquired from 252 sites, of which 51 were pathologically classified as SCC. Optical biomarkers exhibited statistical differences between malignant and normal samples. Contrast was enhanced when parsing tissues by morphological classification rather than anatomical subtype for unpaired comparisons. Corresponding linear discriminant models using multiple optical biomarkers showed improved predictive ability when accounting for morphological classification, particularly in node-positive lesions. The false-positive rate was retrospectively found to decrease by 34.2% in morphologically- vs. anatomically-derived predictive models. In glottic tissue, the surgeon exhibited a false-positive rate of 45.7% while the device showed a lower false-positive rate of 12.4%. Additionally, comparisons of optical parameters were made to further understand the physiology of tumor staging and potential causes of high surgeon false-positive rates. Optical spectroscopy is a user-friendly, non-invasive tool capable of providing quantitative information to discriminate malignant from normal head and neck tissues. Predictive models demonstrated promising results for real-time diagnostics. Furthermore, the strategy described appears to be well suited to reduce the clinical false-positive rate.

Rapid determination of oxygen saturation and vascularity for cancer detection.

A rapid heuristic ratiometric analysis for estimating tissue hemoglobin concentration and oxygen saturation from measured tissue diffuse reflectance spectra is presented. The analysis was validated in tissue-mimicking phantoms and applied to clinical measurements in head and neck, cervical and breast tissues. The analysis works in two steps. First, a linear equation that translates the ratio of the diffuse reflectance at 584 nm and 545 nm to estimate the tissue hemoglobin concentration using a Monte Carlo-based lookup table was developed. This equation is independent of tissue scattering and oxygen saturation. Second, the oxygen saturation was estimated using non-linear logistic equations that translate the ratio of the diffuse reflectance spectra at 539 nm to 545 nm into the tissue oxygen saturation. Correlations coefficients of 0.89 (0.86), 0.77 (0.71) and 0.69 (0.43) were obtained for the tissue hemoglobin concentration (oxygen saturation) values extracted using the full spectral Monte Carlo and the ratiometric analysis, for clinical measurements in head and neck, breast and cervical tissues, respectively. The ratiometric analysis was more than 4000 times faster than the inverse Monte Carlo analysis for estimating tissue hemoglobin concentration and oxygen saturation in simulated phantom experiments. In addition, the discriminatory power of the two analyses was similar. These results show the potential of such empirical tools to rapidly estimate tissue hemoglobin in real-time spectral imaging applications.

Optical and radioiodinated tethered Hsp90 inhibitors reveal selective internalization of ectopic Hsp90 in malignant breast tumor cells.

Inhibitors of heat-shock protein 90 (Hsp90) have demonstrated an unusual selectivity for tumor cells despite its ubiquitous expression. This phenomenon has remained unexplained, but could be influenced by ectopically expressed Hsp90 in tumors. In this work, we synthesized Hsp90 inhibitors that can carry optical or radioiodinated probes via a polyethyleneglycol tether. We show that these tethered inhibitors selectively recognize cells expressing ectopic Hsp90 and become internalized. The internalization process is blocked by Hsp90 antibodies, suggesting that active cycling of the protein occurs at the plasma membrane. In mice, we observed exquisite accumulation of the fluor-tethered versions within breast tumors at very sensitive levels. Cell-based assays with the radiolabeled version showed picomolar detection in cells that express ectopic Hsp90. Our findings show that fluor-tethered or radiolabeled inhibitors that target ectopic Hsp90 can be used to detect breast cancer malignancies through noninvasive imaging.

Compact point-detection fluorescence spectroscopy system for quantifying intrinsic fluorescence redox ratio in brain cancer diagnostics.

We report the development of a compact point-detection fluorescence spectroscopy system and two data analysis methods to quantify the intrinsic fluorescence redox ratio and diagnose brain cancer in an orthotopic brain tumor rat model. Our system employs one compact cw diode laser (407 nm) to excite two primary endogenous fluorophores, reduced nicotinamide adenine dinucleotide, and flavin adenine dinucleotide. The spectra were first analyzed using a spectral filtering modulation method developed previously to derive the intrinsic fluorescence redox ratio, which has the advantages of insensitivity to optical coupling and rapid data acquisition and analysis. This method represents a convenient and rapid alternative for achieving intrinsic fluorescence-based redox measurements as compared to those complicated model-based methods. It is worth noting that the method can also extract total hemoglobin concentration at the same time but only if the emission path length of fluorescence light, which depends on the illumination and collection geometry of the optical probe, is long enough so that the effect of absorption on fluorescence intensity due to hemoglobin is significant. Then a multivariate method was used to statistically classify normal tissues and tumors. Although the first method offers quantitative tissue metabolism information, the second method provides high overall classification accuracy. The two methods provide complementary capabilities for understanding cancer development and noninvasively diagnosing brain cancer. The results of our study suggest that this portable system can be potentially used to demarcate the elusive boundary between a brain tumor and the surrounding normal tissue during surgical resection.