Spatial intratumoral heterogeneity of proliferation in immunohistochemical images of solid tumors

A radiomics-based interpretable model to predict the pathological grade of pancreatic neuroendocrine tumors

OBJECTIVES

To develop a computed tomography (CT) radiomics-based interpretable machine learning (ML) model to predict the pathological grade of pancreatic neuroendocrine tumors (pNETs) in a non-invasive manner.

METHODS

Patients with pNETs who underwent contrast-enhanced abdominal CT between 2010 and 2022 were included in this retrospective study. Radiomics features were extracted, and five radiomics-based ML models, namely logistic regression (LR), random forest (RF), support vector machine (SVM), XGBoost, and GaussianNB, were developed. The performance of these models was evaluated using a time-independent testing set, and metrics such as sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC) were calculated. The accuracy of the radiomics model was compared to that of needle biopsy. The Shapley Additive Explanation (SHAP) tool and the correlation between radiomics and biological features were employed to explore the interpretability of the model.

RESULTS

A total of 122 patients (mean age: 50 ± 14 years; 53 male) were included in the training set, whereas 100 patients (mean age: 48 ± 13 years; 50 male) were included in the testing set. The AUCs for LR, SVM, RF, XGBoost, and GaussianNB were 0.758, 0.742, 0.779, 0.744, and 0.745, respectively, with corresponding accuracies of 73.0%, 70.0%, 77.0%, 71.9%, and 72.9%. The SHAP tool identified two features of the venous phase as the most significant, which showed significant differences among the Ki-67 index or mitotic count subgroups (p < 0.001).

CONCLUSIONS

An interpretable radiomics-based RF model can effectively differentiate between G1 and G2/3 of pNETs, demonstrating favorable interpretability.

CLINICAL RELEVANCE STATEMENT

The radiomics-based interpretable model developed in this study has significant clinical relevance as it offers a non-invasive method for assessing the pathological grade of pancreatic neuroendocrine tumors and holds promise as an important complementary tool to traditional tissue biopsy.

KEY POINTS

• A radiomics-based interpretable model was developed to predict the pathological grade of pNETs and compared with preoperative needle biopsy in terms of accuracy. • The model, based on CT radiomics, demonstrated favorable interpretability. • The radiomics model holds potential as a valuable complementary technique to preoperative needle biopsy; however, it should not be considered a replacement for biopsy.

Interrogating the microenvironmental landscape of tumors with computational image analysis approaches

The tumor microenvironment is an interacting heterogeneous collection of cancer cells, resident as well as infiltrating host cells, secreted factors, and extracellular matrix proteins. With the growing importance of immunotherapies, it has become crucial to be able to characterize the composition and the functional orientation of the microenvironment. The development of novel computational image analysis methodologies may enable the robust quantification and localization of immune and related biomarker-expressing cells within the microenvironment. The aim of the review is to concisely highlight a selection of current and significant contributions pertinent to methodological advances coupled with biomedical or translational applications. A further aim is to concisely present computational advances that, to our knowledge, have currently very limited use for the assessment of the microenvironment but have the potential to enhance image analysis pipelines; on this basis, an example is shown for the detection and segmentation of cells of the microenvironment using a published pipeline and a public dataset. Finally, a general proposal is presented on the conceptual design of automation-optimized computational image analysis workflows in the biomedical and clinical domain.

Detection of Ki67 Hot-Spots of Invasive Breast Cancer Based on Convolutional Neural Networks Applied to Mutual Information of H&E and Ki67 Whole Slide Images

Ki67 hot-spot detection and its evaluation in invasive breast cancer regions play a significant role in routine medical practice. The quantification of cellular proliferation assessed by Ki67 immunohistochemistry is an established prognostic and predictive biomarker that determines the choice of therapeutic protocols. In this paper, we present three deep learning-based approaches to automatically detect and quantify Ki67 hot-spot areas by means of the Ki67 labeling index. To this end, a dataset composed of 100 whole slide images (WSIs) belonging to 50 breast cancer cases (Ki67 and H&E WSI pairs) was used. Three methods based on CNN classification were proposed and compared to create the tumor proliferation map. The best results were obtained by applying the CNN to the mutual information acquired from the color deconvolution of both the Ki67 marker and the H&E WSIs. The overall accuracy of this approach was 95%. The agreement between the automatic Ki67 scoring and the manual analysis is promising with a Spearman’s ρ correlation of 0.92. The results illustrate the suitability of this CNN-based approach for detecting hot-spots areas of invasive breast cancer in WSI.

The Unmet Needs of the Diagnosis, Staging, and Treatment of Gastrointestinal Tumors

New scientific insights in cancer biology and immunobiology have changed the clinical practice of medical oncology in recent years. The molecular stratification of solid tumors has led to improved clinical outcomes and is a key part in the diagnostic workup. Beyond mutational spectra (like Rat sarcoma [RAS] mutations or tumor mutational burden), the investigation of the immunological microenvironment has attracted more efforts. Especially as immunotherapies have changed the standard treatment for some solid tumors dramatically and have become an important part of routine oncology, also for gastrointestinal tumors. Still only a subgroup of patients benefits from immunotherapy in gastrointestinal tumors with prominent examples from colorectal, pancreatic or gastric cancer. Not only microsatellite instability as a marker for response to immunotherapy has shown its utility, there plenty of other approaches currently being investigated to better stratify and understand the microenvironment. But these insights have not translated into clinical utility. Reasons for this are limited technical capabilities for stratification and for coping with heterogeneity of tumor cells and the microenvironment as such. So the situation for gastrointestinal tumors has shown mainly progress for a subgroup of immunotherapy-receptive tumors (eg, microsatellite instability), but advances for the remaining majority have been in the area of stratification and combinatorial therapies, including approaches without chemotherapy. Molecular stratification (eg, B-Rapidly Accelerated Fibrosarcoma [BRAF] V600E mutation in colorectal cancer or NRG1 fusions in Kirsten-rat sarcoma (KRAS) Wild-Type Pancreatic Cancer) has clearly improved the possibilities for directed therapies, but there is a plethora of clinical situations where further developments are needed to improve patient care. Finding these areas and identifying the technical approach to unravel the complexities is the next decisive step. Here the recent advances are summarized and an outlook on possible diagnostic and treatment options in areas of unmet need is given with the context of new molecular imaging possibilities and cutting edge advances in nuclear medicine.

Multilacunarity as a spatial multiscale multi-mass morphometric of change in the meso-architecture of plant parenchyma tissue

The lacunarity index (monolacunarity) averages the behavior of variable size structures in a binary image. The generalized lacunarity concept (multilacunarity) on the basis of generalized distribution moments is an appealing model that can account for differences in the mass content at different scales. The model was tested previously on natural images [J. Vernon-Carter et al., Physica A 388, 4305 (2009)]. Here, the computational aspects of multilacunarity are validated using synthetic binary images that consist of random maps, spatial stochastic patterns, patterns with circular or polygonal elements, and a plane fractal. Furthermore, monolacunarity and detrended fluctuation analysis were employed to quantify the mesostructural changes in the intercellular air spaces of frozen-thawed parenchymatous tissue of pome fruit [N. A. Valous et al., J. Appl. Phys. 115, 064901 (2014)]. Here, the aim is to further examine the coherence of the multilacunarity model for quantifying the mesostructural changes in the intercellular air spaces of parenchymatous tissue of pome and stone fruit, acquired with X-ray microcomputed tomography, after storage and ripening, respectively. The multilacunarity morphometric is a multiscale multi-mass fingerprint of spatial pattern composition, assisting the exploration of the effects of metabolic and physiological activity on the pore space of plant parenchyma tissue.

Variability of the Ki-67 proliferation index in gastroenteropancreatic neuroendocrine neoplasms - a single-center retrospective study

BACKGROUND

The Ki-67 index in gastroenteropancreatic neuroendocrine neoplasms (GEP-NENs) may change throughout the disease course. However, the definitive effect of Ki-67 variability on GEP-NENs remains unknown. The aims of this study were to evaluate changes in Ki-67 levels throughout the disease course and investigate the role of Ki-67 index variability in GEP-NENs.

METHODS

Specimens with multiple pathologies were evaluated from 30 patients who were selected from 514 patients with GEP-NENs, being treated at Wuhan Union Hospital from July 2009 to February 2018. The Ki-67 index was evaluated among multiple specimens over the disease course. Univariable and multivariable Cox proportional hazards regression analyses were performed to assess the prognostic significance of various clinical and histopathologic features.

RESULTS

Among the 514 patients with GEP-NENs, metastases were seen in 182 (35.41%). Among the 30 patients from whom specimens with multiple pathologies were obtained, 24 were both primary and metastatic specimens and six were specimens collected over the course of the disease. Changes in Ki-67 levels were detected in 53.3% of the patients, of whom 40% had up-regulated Ki-67 levels, and 13.3% had down-regulated Ki-67 levels. Kaplan-Meier survival analysis showed that the group with Ki-67 variability had a shorter overall survival (p = 0.0297). The Cox regression analysis indicated that Ki-67 variability (p = 0.038) was the only independent prognostic factor for overall survival.

CONCLUSIONS

Our data suggest that patients with GEP-NENs and Ki-67 variability had a poorer prognosis. The re-assessment of Ki-67 at sites of metastasis or during the disease course might play a role in predicting the prognosis of patients with GEP-NENs. This finding could have implications for how GEP-NENs are monitored and treated.

Analysis of the proliferative activity in lung adenocarcinomas with specific driver mutations

In the last decade it became evident that many lung adenocarcinomas (ADC) harbor key genetic alterations such as KRAS, EGFR or BRAF mutations as well as rearrangements of ROS1 or ALK that drive these tumors. In the present study we investigated whether different driver mutations of ADC result in different proliferation rates, which might have clinical impact, including resistance to therapy, recurrence and prognosis. We analyzed the proliferation index (PI) on full slides of surgically resected ADC (n = 230) with known genetic aberrations by means of immunohistochemistry and subsequent digital image analysis and correlated the results with clinicopathological variables including overall (OS) and disease free survival (DFS). We did not observe significant differences in OS or DFS regarding the KRAS or EGFR mutational status (P = 0.56). However, KRAS mutated ADC showed an increased PI compared to EGFR mutated ADC, and ADC with ALK translocations (P < 0.01). Subgroup analysis of EGFR mutated ADC showed a higher PI for tumors harboring a mutation in exon 18 and 20, compared to tumors with a mutation in exon 19 or 21. A PI of 11.5% was the best possible prognostic stratificator for OS (P = 0.01 in KRAS mutated and P < 0.01 in EGFR mutated ADC). In conclusion, the PI differs significantly among ADC with distinct driver mutations. This might explain the varying indications for a prognostic relevance of the PI observed in prior studies. Our study provides a basis for the establishment of a reliable and clinically meaningful PI threshold.

Robust detection and segmentation of cell nuclei in biomedical images based on a computational topology framework

The segmentation of cell nuclei is an important step towards the automated analysis of histological images. The presence of a large number of nuclei in whole-slide images necessitates methods that are computationally tractable in addition to being effective. In this work, a method is developed for the robust segmentation of cell nuclei in histological images based on the principles of persistent homology. More specifically, an abstract simplicial homology approach for image segmentation is established. Essentially, the approach deals with the persistence of disconnected sets in the image, thus identifying salient regions that express patterns of persistence. By introducing an image representation based on topological features, the task of segmentation is less dependent on variations of color or texture. This results in a novel approach that generalizes well and provides stable performance. The method conceptualizes regions of interest (cell nuclei) pertinent to their topological features in a successful manner. The time cost of the proposed approach is lower-bounded by an almost linear behavior and upper-bounded by O(n²) in a worst-case scenario. Time complexity matches a quasilinear behavior which is O(n^1+ɛ) for ε < 1. Images acquired from histological sections of liver tissue are used as a case study to demonstrate the effectiveness of the approach. The histological landscape consists of hepatocytes and non-parenchymal cells. The accuracy of the proposed methodology is verified against an automated workflow created by the output of a conventional filter bank (validated by experts) and the supervised training of a random forest classifier. The results are obtained on a per-object basis. The proposed workflow successfully detected both hepatocyte and non-parenchymal cell nuclei with an accuracy of 84.6%, and hepatocyte cell nuclei only with an accuracy of 86.2%. A public histological dataset with supplied ground-truth data is also used for evaluating the performance of the proposed approach (accuracy: 94.5%). Further validations are carried out with a publicly available dataset and ground-truth data from the Gland Segmentation in Colon Histology Images Challenge (GlaS) contest. The proposed method is useful for obtaining unsupervised robust initial segmentations that can be further integrated in image/data processing and management pipelines. The development of a fully automated system supporting a human expert provides tangible benefits in the context of clinical decision-making.

Detailed resolution analysis reveals spatial T cell heterogeneity in the invasive margin of colorectal cancer liver metastases associated with improved survival

On a broader scale, T cell density and localization in colorectal cancer liver metastases have prognostic and predictive implications. As T cell distribution at higher resolutions has not been fully investigated, a detailed resolution analysis of T cell distribution was performed. Patient tissues were divided into 10 µm distance classes between the tumor border and adjacent normal liver. Thereby, distinct density patterns of T cell localization in relation to the malignant tissue could be detected. At a distance of 20 to 30 µm to the tumor, a decrease of CD3 T cells is common. Within this area, cytotoxic Granzyme B and CD8⁺ T cells were found to be significantly reduced as well as CD163 macrophages were increased and identified to be in close contact with T cells. Our data suggests a physical or functional border within this region. Survival analysis revealed improved overall survival in patients with high T cells numbers at the direct tumor border. Interestingly, the decreased T cells in the 20 to 30 µm region were also found to be significantly associated with improved survival. Consequently, the detailed localization of T cells, despite blockade, could be associated with improved clinical outcome. The high-resolution analysis represents new insights into relevant heterogenous T cell distributions especially related to clinical responses. As the paradoxical observation of localization-dependent prognostic relevance of T cell densities is only detectable by detailed spatial analyses, this investigation of spatial profiles at higher resolutions is suggested as a new biomarker for survival and response to therapies.