Digital Histopathological Discrimination of Label-Free Tumoral Tissues by Artificial Intelligence Phase-Imaging Microscopy.

Histopathology is the gold standard for disease diagnosis. The use of digital histology on fresh samples can reduce processing time and potential image artifacts, as label-free samples do not need to be fixed nor stained. This fact allows for a faster diagnosis, increasing the speed of the process and the impact on patient prognosis. This work proposes, implements, and validates a novel digital diagnosis procedure of fresh label-free histological samples. The procedure is based on advanced phase-imaging microscopy parameters and artificial intelligence. Fresh human histological samples of healthy and tumoral liver, kidney, ganglion, testicle and brain were collected and imaged with phase-imaging microscopy. Advanced phase parameters were calculated from the images. The statistical significance of each parameter for each tissue type was evaluated at different magnifications of 10×, 20× and 40×. Several classification algorithms based on artificial intelligence were applied and evaluated. Artificial Neural Network and Decision Tree approaches provided the best general sensibility and specificity results, with values over 90% for the majority of biological tissues at some magnifications. These results show the potential to provide a label-free automatic significant diagnosis of fresh histological samples with advanced parameters of phase-imaging microscopy. This approach can complement the present clinical procedures.

Software-assisted decision support in digital histopathology.

Tissue diagnostics is the world of pathologists, and it is increasingly becoming digitalised to leverage the enormous potential of personalised medicine and of stratifying patients, enabling the administration of modern therapies. Therefore, the daily task for pathologists is changing drastically and will become increasingly demanding in order to take advantage of the development of modern computer technologies. The role of pathologist has rapidly evolved from exclusively describing the morphology and phenomenology of a disease, to becoming a gatekeeper for novel and most effective treatment options. This is possible based on the retrieval and management of a wide range of complex information from tissue or a group of cells and associated meta-data. Intelligent and self-learning software solutions can support and guide pathologists to score clinically relevant decisions based on the accurate and robust quantification of multiple target molecules or surrogate biomarker as companion or complimentary diagnostics along with relevant spatial relationships and contextual information from digital H&E and multiplexed images. With the availability of multiplex staining techniques on a single slide, high-resolution image analysis tools, and high-end computer hardware, machine and deep learning solutions now offer diagnostic rulesets and algorithms that still require clinical validation in well-designed studies. Before entering the clinical practice, the 'human factor' pathologist needs to develop trust in the output coming from the 'digital black box of computational pathology', including image analysis solutions and artificial intelligence algorithms to support critical clinical decisions which otherwise would not be available. © 2020 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

Recent Advances in Pathology: the 2020 Annual Review Issue of The Journal of Pathology.

This year's Annual Review Issue of The Journal of Pathology contains 18 invited reviews on current research areas in pathology. The subject areas reflect the broad range of topics covered by the journal and this year encompass the development and application of software in digital histopathology, implementation of biomarkers in pathology practice; genetics and epigenetics, and stromal influences in disease. The reviews are authored by experts in their field and provide comprehensive updates in the chosen areas, in which there has been considerable recent progress in our understanding of disease. © 2020 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

Shedding Light on Metal-Based Nanoparticles in Zebrafish by Computed Tomography with Micrometer Resolution.

Metal-based nanoparticles are clinically used for diagnostic and therapeutic applications. After parenteral administration, they will distribute throughout different organs. Quantification of their distribution within tissues in the 3D space, however, remains a challenge owing to the small particle diameter. In this study, synchrotron radiation-based hard X-ray tomography (SRµCT) in absorption and phase contrast modes is evaluated for the localization of superparamagnetic iron oxide nanoparticles (SPIONs) in soft tissues based on their electron density and X-ray attenuation. Biodistribution of SPIONs is studied using zebrafish embryos as a vertebrate screening model. This label-free approach gives rise to an isotropic, 3D, direct space visualization of the entire 2.5 mm-long animal with a spatial resolution of around 2 µm. High resolution image stacks are available on a dedicated internet page (http://zebrafish.pharma-te.ch). X-ray tomography is combined with physico-chemical characterization and cellular uptake studies to confirm the safety and effectiveness of protective SPION coatings. It is demonstrated that SRµCT provides unprecedented insights into the zebrafish embryo anatomy and tissue distribution of label-free metal oxide nanoparticles.

MALDI imaging mass spectrometry: discrimination of pathophysiological regions in traumatized skeletal muscle by characteristic peptide signatures.

Due to formation of fibrosis and the loss of contractile muscle tissue, severe muscle injuries often result in insufficient healing marked by a significant reduction of muscle force and motor activity. Our previous studies demonstrated that the local transplantation of mesenchymal stromal cells into an injured skeletal muscle of the rat improves the functional outcome of the healing process. Since, due to the lack of sufficient markers, the accurate discrimination of pathophysiological regions in injured skeletal muscle is inadequate, underlying mechanisms of the beneficial effects of mesenchymal stromal cell transplantation on primary trauma and trauma adjacent muscle area remain elusive. For discrimination of these pathophysiological regions, formalin-fixed injured skeletal muscle tissue was analyzed by MALDI imaging MS. By using two computational evaluation strategies, a supervised approach (ClinProTools) and unsupervised segmentation (SCiLS Lab), characteristic m/z species could be assigned to primary trauma and trauma adjacent muscle regions. Using "bottom-up" MS for protein identification and validation of results by immunohistochemistry, we could identify two proteins, skeletal muscle alpha actin and carbonic anhydrase III, which discriminate between the secondary damage on adjacent tissue and the primary traumatized muscle area. Our results underscore the high potential of MALDI imaging MS to describe the spatial characteristics of pathophysiological changes in muscle.

Deep Learning Analysis for Predicting Tumor Spread through Air Space in Early-Stage Lung Adenocarcinoma Pathology Images.

The presence of spread through air spaces (STASs) in early-stage lung adenocarcinoma is a significant prognostic factor associated with disease recurrence and poor outcomes. Although current STAS detection methods rely on pathological examinations, the advent of artificial intelligence (AI) offers opportunities for automated histopathological image analysis. This study developed a deep learning (DL) model for STAS prediction and investigated the correlation between the prediction results and patient outcomes. To develop the DL-based STAS prediction model, 1053 digital pathology whole-slide images (WSIs) from the competition dataset were enrolled in the training set, and 227 WSIs from the National Taiwan University Hospital were enrolled for external validation. A YOLOv5-based framework comprising preprocessing, candidate detection, false-positive reduction, and patient-based prediction was proposed for STAS prediction. The model achieved an area under the curve (AUC) of 0.83 in predicting STAS presence, with 72% accuracy, 81% sensitivity, and 63% specificity. Additionally, the DL model demonstrated a prognostic value in disease-free survival compared to that of pathological evaluation. These findings suggest that DL-based STAS prediction could serve as an adjunctive screening tool and facilitate clinical decision-making in patients with early-stage lung adenocarcinoma.

Integration of virtual microscopy podcasts in the histology discipline in osteopathic medical school: Learning outcomes.

Virtual microscopy podcasts (VMPs) are narrative recordings of digital histology images. This study evaluated the outcomes of integrating the VMPs into teaching histology to osteopathic medical students. The hypothesis was that incorporating virtual microscopy podcasts as supplementary histology resources to the curriculum would have a positive impact on student performance and satisfaction. Sixty-one podcasts of dynamic microscopic images were created using screen recordings of the digital slides. The VMPs were integrated as supplementary histology resources in multiple courses during the first and second years of the medical curriculum for three classes, a total of 477 osteopathic medical students. A voluntary and anonymous survey was obtained from the students using a questionnaire that included two open-ended questions. The overall performance of the three classes on the histology content of the preclinical course examinations was compared to historical controls of the previous two classes that did not have access to the VMPs. Most students indicated that the podcasts enabled more efficient study time and improved their confidence in the histology content on examinations. The findings indicated a positive association between podcast viewing and efficient study time utilization and class performance. The class average scores of the three consecutive cohorts that used the VMPs progressively increased by 7.69%, 14.88%, and 14.91% compared to the controls. A summary of students' feedback and academic performance supported that integration of the VMPs into Histology teaching improved the learning experience. The findings align with previous studies on the effectiveness of multimedia-based teaching in histology laboratory modules.

Webcam-based eye-tracking to measure visual expertise of medical students during online histology training.

Objectives: Visual expertise is essential for image-based tasks that rely on visual cues, such as in radiology or histology. Studies suggest that eye movements are related to visual expertise and can be measured by near-infrared eye-tracking. With the popularity of device-embedded webcam eye-tracking technology, cost-effective use in educational contexts has recently become amenable. This study investigated the feasibility of such methodology in a curricular online-only histology course during the 2021 summer term. Methods: At two timepoints (t1 and t2), third-semester medical students were asked to diagnose a series of histological slides while their eye movements were recorded. Students' eye metrics, performance and behavioral measures were analyzed using variance analyses and multiple regression models. Results: First, webcam-eye tracking provided eye movement data with satisfactory quality (mean accuracy=115.7 px±31.1). Second, the eye movement metrics reflected the students' proficiency in finding relevant image sections (fixation count on relevant areas=6.96±1.56 vs. irrelevant areas=4.50±1.25). Third, students' eye movement metrics successfully predicted their performance (R2adj=0.39, p<0.001). Conclusion: This study supports the use of webcam-eye-tracking expanding the range of educational tools available in the (digital) classroom. As the students' interest in using the webcam eye-tracking was high, possible areas of implementation will be discussed.

Quantitative analysis of breast cancer tissue composition and associations with tumor subtype.

The tumor microenvironment is an important determinant of breast cancer progression, but standard methods for describing the tumor microenvironment are lacking. Measures of microenvironment composition such as stromal area and immune infiltrate are labor-intensive and few large studies have systematically collected this data. However, digital histologic approaches are becoming more widely available, allowing high-throughput, quantitative estimation. We applied such methods to tissue microarrays of tumors from 1687 women (mean 4 cores per case) in the Carolina Breast Cancer Study Phase 3. Tumor composition was quantified as percentage of epithelium, stroma, adipose, and lymphocytic infiltrate (with the latter as presence/absence using a ≥1% cutoff). Composition proportions and presence/absence were evaluated in association with clinical and molecular features of breast cancer (intrinsic subtype and RNA-based risk of recurrence [ROR] scores) using multivariable linear and logistic regression. Lower stromal content was associated with aggressive tumor phenotypes, including triple-negative (31.1% vs. 41.6% in HR+/HER2-; RFD [95% CI]: -10.5%, [-13.1, -7.9]), Basal-like subtypes (29.0% vs. 44.0% in Luminal A; RFD [95% CI]: -14.9%, [-17.8, -12.0]), and high RNA-based PAM50 ROR scores (27.6% vs. 48.1% in ROR low; RFD [95% CI]: -20.5%, [24.3, 16.7]), after adjusting for age and race. HER2+ tumors also had lower stromal content, particularly among RNA-based HER2-enriched (35.2% vs. 44.0% in Luminal A; RFD [95% CI]: -8.8%, [-13.8, -3.8]). Similar associations were observed between immune infiltrate and tumor phenotypes. Quantitative digital image analysis of the breast cancer microenvironment showed significant associations with demographic characteristics and biological indicators of aggressive behavior.

Digital Histological Study of Neocortical Grey and White Matter Tau Burden Across Tauopathies.

3R/4R-tau species are found in Alzheimer disease (AD) and ∼50% of Lewy body dementias at autopsy (LBD+tau); 4R-tau accumulations are found in progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD). Digital image analysis techniques can elucidate patterns of tau pathology more precisely than traditional methods but repeatability across centers is unclear. We calculated regional percentage areas occupied by tau pathological inclusions from the middle frontal cortex (MFC), superior temporal cortex (STC), and angular gyrus (ANG) from cases from the University of Pennsylvania and the University of California San Diego with AD, LBD+tau, PSP, or CBD (n = 150) using QuPath. In both cohorts, AD and LBD+tau had the highest grey and white matter tau burden in the STC (p ≤ 0.04). White matter tau burden was relatively higher in 4R-tauopathies than 3R/4R-tauopathies (p < 0.003). Grey and white matter tau were correlated in all diseases (R2=0.43-0.79, p < 0.04) with the greatest increase of white matter per unit grey matter tau observed in PSP (p < 0.02 both cohorts). Grey matter tau negatively correlated with MMSE in AD and LBD+tau (r = -4.4 to -5.4, p ≤ 0.02). These data demonstrate the feasibility of cross-institutional digital histology studies that generate finely grained measurements of pathology which can be used to support biomarker development and models of disease progression.

Toward a digital analysis of environmental impacts on rodent mammary gland density during critical developmental windows.

While mammographic breast density is associated with breast cancer risk in humans, there is no comparable surrogate risk measure in mouse and rat mammary glands following various environmental exposures. In the current study, mammary glands from mice and rats subjected to reproductive factors and exposures to environmental chemicals that have been shown to influence mammary gland development and/or susceptibility to mammary tumors were evaluated for histologic density by manual and automated digital methods. Digital histological density detected changes due to hormonal stimuli/reproductive factors (parity), dietary fat, and exposure to environmental chemicals, such as benzophenone-3 and a combination of perfluorooctanoic acid and zeranol. Thus, digital analysis of mammary gland density offers a high throughput method that can provide a highly reproducible means of comparing a measure of histological density across independent experiments, experimental systems, and laboratories. This methodology holds promise for the detection of environmental impacts on mammary gland structure in mice and rats that may be comparable to human breast density, thus potentially allowing comparisons between rodent models and human breast cancer studies.

Implementation of a fully digital histology course in the anatomical teaching curriculum during COVID-19 pandemic.

BACKGROUND: During the COVID-19 pandemic, many medical schools are forced to switch courses of the mandatory curriculum to online teaching formats. However, little information about feasibility and effectiveness is available yet about distance teaching in anatomy. The aim of this study was to evaluate the implementation of a histology course previously taught in a classroom setting into an online-only format based on video conference software. METHODS: Our course design included theoretical introductions, an online-adaptation of virtual microscopy used previously in the classroom, and active learning elements such as collaborative learning in breakout rooms, annotation assignments and multiple-choice questions. Two preclinical semester cohorts of around 400 second and third semester students were taught in histology in parallel courses, using the Zoom software platform. We analyzed data about student attendance during the course, summative quantitative and qualitative evaluation of the students and results of a written test required to pass the course. RESULTS: We observed that student attendance was high and stable during the 19 course days for both second and third semester, and only few students reported technical problems. There were no significant differences in examination results of second semester compared to the third semester, an unexpected result as the third semester already participated in the dissection course before. Similarly, no significant gender-related effects on the examination performance could be noted in both semesters. However, the age of students was negatively correlated with test scores in the second and third semester. Importantly, the overall evaluation of the digital version of the histology course was at least as positive as the in-person version over the past years. CONCLUSION: Together, we experienced that the implementation of a curricular histology course in an online-format is technically realizable, effective and well accepted among students. We also observed that availability and prior experience with digitized specimen in virtual microscopy facilitates transition into an online-only setting. Thus, our study supports the positive potential of distance learning for teaching anatomy during and after COVID-19 pandemic but also emphasizes the need for a synchronous learning environment with partially personnel-intensive small group settings to overcome passivity and inequality aspects, and to foster active learning elements.

Aquaporin-4 Expression during Toxic and Autoimmune Demyelination.

The water channel protein aquaporin-4 (AQP4) is required for a normal rate of water exchange across the blood-brain interface. Following the discovery that AQP4 is a possible autoantigen in neuromyelitis optica, the function of AQP4 in health and disease has become a research focus. While several studies have addressed the expression and function of AQP4 during inflammatory demyelination, relatively little is known about its expression during non-autoimmune-mediated myelin damage. In this study, we used the toxin-induced demyelination model cuprizone as well as a combination of metabolic and autoimmune myelin injury (i.e., Cup/EAE) to investigate AQP4 pathology. We show that during toxin-induced demyelination, diffuse AQP4 expression increases, while polarized AQP4 expression at the astrocyte endfeet decreases. The diffuse increased expression of AQP4 was verified in chronic-active multiple sclerosis lesions. Around inflammatory brain lesions, AQP4 expression dramatically decreased, especially at sites where peripheral immune cells penetrate the brain parenchyma. Humoral immune responses appear not to be involved in this process since no anti-AQP4 antibodies were detected in the serum of the experimental mice. We provide strong evidence that the diffuse increase in anti-AQP4 staining intensity is due to a metabolic injury to the brain, whereas the focal, perivascular loss of anti-AQP4 immunoreactivity is mediated by peripheral immune cells.

Digital histology with Mueller microscopy: how to mitigate an impact of tissue cut thickness fluctuations.

Mueller microscopy studies of fixed unstained histological cuts of human skin models were combined with an analysis of experimental data within the framework of differential Mueller matrix (MM) formalism. A custom-built Mueller polarimetric microscope was used in transmission configuration for the optical measurements of skin tissue model adjacent cuts of various nominal thicknesses (5 to 30 µm). The maps of both depolarization and polarization parameters were calculated from the corresponding microscopic MM images by applying a logarithmic Mueller matrix decomposition (LMMD) pixelwise. The parameters derived from LMMD of measured tissue cuts and the intensity of transmitted light were used for an automated segmentation of microscopy images to delineate dermal and epidermal layers. The quadratic dependence of depolarization parameters and linear dependence of polarization parameters on thickness, as predicted by the theory, was confirmed in our measurements. These findings pave the way toward digital histology with polarized light by presenting the combination of optimal optical markers, which allows mitigating the impact of tissue cut thickness fluctuations and increases the contrast of polarimetric images for tissue diagnostics.

Vascular density of histologically benign breast tissue from women with breast cancer: associations with tissue composition and tumor characteristics.

In breast tumors, it is well established that intratumoral angiogenesis is crucial for malignant progression, but little is known about the vascular characteristics of extratumoral, cancer-adjacent breast. Genome-wide transcriptional data suggest that extratumoral microenvironments may influence breast cancer phenotypes; thus, histologic features of cancer-adjacent tissue may also have clinical implications. To this end, we developed a digital algorithm to quantitate vascular density in approximately 300 histologically benign tissue specimens from breast cancer patients enrolled in the UNC Normal Breast Study (NBS). Specimens were stained for CD31, and vascular content was compared to demographic variables, tissue composition metrics, and tumor molecular features. We observed that the vascular density of cancer-adjacent breast was significantly higher in older and obese women, and was strongly associated with breast adipose tissue content. Consistent with observations that older and heavier women experience higher frequencies of ER+ disease, higher extratumoral vessel density was also significantly associated with positive prognostic tumor features such as lower stage, negative nodal status, and smaller size (<2 cm). These results reveal biological relationships between extratumoral vascular content and body size, breast tissue composition, and tumor characteristics, and suggest biological plausibility for the relationship between weight gain (and corresponding breast tissue changes) and breast cancer progression.

An Expert Derived Feedforward Histology Module Improves Pattern Recognition Efficiency in Novice Students.

Histology is a visually oriented, foundational anatomical sciences subject in professional health curricula that has seen a dramatic reduction in educational contact hours and an increase in content migration to a digital platform. While the digital migration of histology laboratories has transformed histology education, few studies have shown the impact of this change on visual literacy development, a critical competency in histology. The objective of this study was to assess whether providing a video clip of an expert's gaze while completing leukocyte identification tasks would increase the efficiency and performance of novices completing similar identification tasks. In a randomized study, one group of novices (n = 9) was provided with training materials that included expert eye gaze, while the other group (n = 12) was provided training materials with identical content, but without the expert eye gaze. Eye movement parameters including fixation rate and total scan path distance, and performance measures including time-to-task-completion and accuracy, were collected during an identification task assessment. Compared to the control group, the average fixation duration was 13.2% higher (P < 0.02) and scan path distance was 35.0% shorter in the experimental group (P = 0.14). Analysis of task performance measures revealed no significant difference between the groups. These preliminary results suggest a more efficient search performed by the experimental group, indicating the potential efficacy of training using an expert's gaze to enhance visual literacy development. With further investigation, such feedforward enhanced training methods could be utilized for histology and other visually oriented subjects.

Deformable Registration of Histological Cancer Margins to Gross Hyperspectral Images using Demons.

Hyperspectral imaging (HSI), a non-contact optical imaging technique, has been recently used along with machine learning technique to provide diagnostic information about ex-vivo surgical specimens for optical biopsy. The computer-aided diagnostic approach requires accurate ground truths for both training and validation. This study details a processing pipeline for registering the cancer-normal margin from a digitized histological image to the gross-level HSI of a tissue specimen. Our work incorporates an initial affine and control-point registration followed by a deformable Demons-based registration of the moving mask obtained from the histological image to the fixed mask made from the HS image. To assess registration quality, Dice similarity coefficient (DSC) measures the image overlap, visual inspection is used to evaluate the margin, and average target registration error (TRE) of needle-bored holes measures the registration error between the histologic and HSI images. Excised tissue samples from seventeen patients, 11 head and neck squamous cell carcinoma (HNSCCa) and 6 thyroid carcinoma, were registered according to the proposed method. Three registered specimens are illustrated in this paper, which demonstrate the efficacy of the registration workflow. Further work is required to apply the technique to more patient data and investigate the ability of this procedure to produce suitable gold standards for machine learning validation.

Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases.

BACKGROUND: Deep learning (DL) is a representation learning approach ideally suited for image analysis challenges in digital pathology (DP). The variety of image analysis tasks in the context of DP includes detection and counting (e.g., mitotic events), segmentation (e.g., nuclei), and tissue classification (e.g., cancerous vs. non-cancerous). Unfortunately, issues with slide preparation, variations in staining and scanning across sites, and vendor platforms, as well as biological variance, such as the presentation of different grades of disease, make these image analysis tasks particularly challenging. Traditional approaches, wherein domain-specific cues are manually identified and developed into task-specific "handcrafted" features, can require extensive tuning to accommodate these variances. However, DL takes a more domain agnostic approach combining both feature discovery and implementation to maximally discriminate between the classes of interest. While DL approaches have performed well in a few DP related image analysis tasks, such as detection and tissue classification, the currently available open source tools and tutorials do not provide guidance on challenges such as (a) selecting appropriate magnification, (b) managing errors in annotations in the training (or learning) dataset, and (c) identifying a suitable training set containing information rich exemplars. These foundational concepts, which are needed to successfully translate the DL paradigm to DP tasks, are non-trivial for (i) DL experts with minimal digital histology experience, and (ii) DP and image processing experts with minimal DL experience, to derive on their own, thus meriting a dedicated tutorial. AIMS: This paper investigates these concepts through seven unique DP tasks as use cases to elucidate techniques needed to produce comparable, and in many cases, superior to results from the state-of-the-art hand-crafted feature-based classification approaches. RESULTS: Specifically, in this tutorial on DL for DP image analysis, we show how an open source framework (Caffe), with a singular network architecture, can be used to address: (a) nuclei segmentation (F-score of 0.83 across 12,000 nuclei), (b) epithelium segmentation (F-score of 0.84 across 1735 regions), (c) tubule segmentation (F-score of 0.83 from 795 tubules), (d) lymphocyte detection (F-score of 0.90 across 3064 lymphocytes), (e) mitosis detection (F-score of 0.53 across 550 mitotic events), (f) invasive ductal carcinoma detection (F-score of 0.7648 on 50 k testing patches), and (g) lymphoma classification (classification accuracy of 0.97 across 374 images). CONCLUSION: This paper represents the largest comprehensive study of DL approaches in DP to date, with over 1200 DP images used during evaluation. The supplemental online material that accompanies this paper consists of step-by-step instructions for the usage of the supplied source code, trained models, and input data.

Histomolecular interpretation of pleomorphic adenomas of the salivary gland by matrix-assisted laser desorption ionization imaging and spatial segmentation.

BACKGROUND: Despite efforts in localization of key proteins using immunohistochemistry, the complex proteomic composition of pleomorphic adenomas has not yet been characterized. Matrix-assisted laser desorption/ionization imaging mass spectrometry (MALDI imaging) allows label-free and spatially resolved detection of hundreds of proteins directly from tissue sections and of histomorphological regions by finding colocalized molecular signals. Spatial segmentation of MALDI imaging data is an algorithmic method for finding regions of similar proteomic composition as functionally similar regions. METHODS: We investigated 2 pleomorphic adenomas by applying spatial segmentation to the MALDI imaging data of tissue sections. RESULTS: The spatial segmentation subdivided the tissue in a good accordance with the tissue histology. Numerous molecular signals colocalized with histologically defined tissue regions were found. CONCLUSION: Our study highlights the cellular transdifferentiation within the pleomorphic adenoma. It could be shown that spatial segmentation of MALDI imaging data is a promising approach in the emerging field of digital histological analysis and characterization of tumors.

Computer-aided identification of prostatic adenocarcinoma: Segmentation of glandular structures.

BACKGROUND: Identification of individual prostatic glandular structures is an important prerequisite to quantitative histological analysis of prostate cancer with the aid of a computer. We have developed a computer method to segment individual glandular units and to extract quantitative image features, for computer identification of prostatic adenocarcinoma. METHODS: TWO SETS OF DIGITAL HISTOLOGY IMAGES WERE USED: database I (n = 57) for developing and testing the computer technique, and database II (n = 116) for independent validation. The segmentation technique was based on a k-means clustering and a region-growing method. Computer segmentation results were evaluated subjectively and also compared quantitatively against manual gland outlines, using the Jaccard similarity measure. Quantitative features that were extracted from the computer segmentation results include average gland size, spatial gland density, and average gland circularity. Linear discriminant analysis (LDA) was used to combine quantitative image features. Classification performance was evaluated with receiver operating characteristic (ROC) analysis and the area under the ROC curve (AUC). RESULTS: Jaccard similarity coefficients between computer segmentation and manual outlines of individual glands were between 0.63 and 0.72 for non-cancer and between 0.48 and 0.54 for malignant glands, respectively, similar to an interobserver agreement of 0.79 for non-cancer and 0.75 for malignant glands, respectively. The AUC value for the features of average gland size and gland density combined via LDA was 0.91 for database I and 0.96 for database II. CONCLUSIONS: Using a computer, we are able to delineate individual prostatic glands automatically and identify prostatic adenocarcinoma accurately, based on the quantitative image features extracted from computer-segmented glandular structures.