Ouroboros: cross-linking protein expression perturbations and cancer histology imaging with generative-predictive modeling.

SUMMARY: Imagine if we could simultaneously predict spatial protein expression in tissues from their routine Hematoxylin and Eosin (H&E) stained images, and create tissue images given protein expression profiles thus enabling virtual simulations of how protein expression alterations impact histology in complex diseases like cancer. Such an approach could lead to more informed diagnostic and therapeutic decisions for precision medicine at lower costs and shorter turnaround times, more detailed insights into underlying disease pathology as well as improvement in predictive and generative performance. In this study, we investigate the intricate correlation between protein expressions obtained from Hyperion mass cytometry and histopathological microstructures in conventional H&E stained glioblastoma (GBM) samples, unveiling morphological patterns and cellular-level spatial alterations associated with protein expression changes. To model these complex relationships, we propose a novel generative-predictive framework called Ouroboros for producing H&E images from protein expressions and simultaneously predicting protein expressions from H&E images. Our comprehensive sample-independent validation over 9920 tissue spots from 4 GBM samples encompassing visual image analysis, quantitative analysis, subspace alignment and perturbation experiments shows that the proposed generative-predictive approach offers significant improvements in predicting protein expression from images in comparison to baseline methods as well as accurate generation of virtual GBM sample images. This proof of concept study can contribute to advancing our understanding of histological responses to protein expression perturbations and lays the foundations for further developments in this area. AVAILABILITY AND IMPLEMENTATION: Implementation and associated data for the proposed approach are available at the URL: https://github.com/Srijay/Ouroboros.

Integrating Spatial and Morphological Characteristics into Melanoma Prognosis: A Computational Approach.

In this study, the prognostic value of cellular morphology and spatial configurations in melanoma has been examined, aiming to complement traditional prognostic indicators like mitotic activity and tumor thickness. Through a computational pipeline using machine learning and deep learning methods, we quantified nuclei sizes within different spatial regions and analyzed their prognostic significance using univariate and multivariate Cox models. Nuclei sizes in the invasive band demonstrated a significant hazard ratio (HR) of 1.1 (95% CI: 1.03, 1.18). Similarly, the nuclei sizes of tumor cells and Ki67 S100 co-positive cells in the invasive band achieved HRs of 1.07 (95% CI: 1.02, 1.13) and 1.09 (95% CI: 1.04, 1.16), respectively. Our findings reveal that nuclei sizes, particularly in the invasive band, are potentially prognostic factors. Correlation analyses further demonstrated a meaningful relationship between cellular morphology and tumor progression, notably showing that nuclei size within the invasive band correlates substantially with tumor thickness. These results suggest the potential of integrating spatial and morphological analyses into melanoma prognostication.

Artificial Intelligence for Image-Based Breast Cancer Risk Prediction Using Attention.

Accurate prediction of individual breast cancer risk paves the way for personalised prevention and early detection. The incorporation of genetic information and breast density has been shown to improve predictions for existing models, but detailed image-based features are yet to be included despite correlating with risk. Complex information can be extracted from mammograms using deep-learning algorithms, however, this is a challenging area of research, partly due to the lack of data within the field, and partly due to the computational burden. We propose an attention-based Multiple Instance Learning (MIL) model that can make accurate, short-term risk predictions from mammograms taken prior to the detection of cancer at full resolution. Current screen-detected cancers are mixed in with priors during model development to promote the detection of features associated with risk specifically and features associated with cancer formation, in addition to alleviating data scarcity issues. MAI-risk achieves an AUC of 0.747 [0.711, 0.783] in cancer-free screening mammograms of women who went on to develop a screen-detected or interval cancer between 5 and 55 months, outperforming both IBIS (AUC 0.594 [0.557, 0.633]) and VAS (AUC 0.649 [0.614, 0.683]) alone when accounting for established clinical risk factors.

Lymphoma triage from H&E using AI for improved clinical management.

AIMS: In routine diagnosis of lymphoma, initial non-specialist triage is carried out when the sample is biopsied to determine if referral to specialised haematopathology services is needed. This places a heavy burden on pathology services, causes delays and often results in over-referral of benign cases. We aimed to develop an automated triage system using artificial intelligence (AI) to enable more accurate and rapid referral of cases, thereby addressing these issues. METHODS: A retrospective dataset of H&E-stained whole slide images (WSI) of lymph nodes was taken from Newcastle University Hospital (302 cases) and Manchester Royal Infirmary Hospital (339 cases) with approximately equal representation of the 3 most prevalent lymphoma subtypes: follicular lymphoma, diffuse large B-cell and classic Hodgkin's lymphoma, as well as reactive controls. A subset (80%) of the data was used for training, a further validation subset (10%) for model selection and a final non-overlapping test subset (10%) for clinical evaluation. RESULTS: AI triage achieved multiclass accuracy of 0.828±0.041 and overall accuracy of 0.932±0.024 when discriminating between reactive and malignant cases. Its ability to detect lymphoma was equivalent to that of two haematopathologists (0.925, 0.950) and higher than a non-specialist pathologist (0.75) repeating the same task. To aid explainability, the AI tool also provides uncertainty estimation and attention heatmaps. CONCLUSIONS: Automated triage using AI holds great promise in contributing to the accurate and timely diagnosis of lymphoma, ultimately benefiting patient care and outcomes.

Spatial and molecular profiling of the mononuclear phagocyte network in classic Hodgkin lymphoma.

Classic Hodgkin lymphoma (cHL) has a rich immune infiltrate, which is an intrinsic component of the neoplastic process. Malignant Hodgkin Reed-Sternberg cells (HRSCs) create an immunosuppressive microenvironment by the expression of regulatory molecules, preventing T-cell activation. It has also been demonstrated that mononuclear phagocytes (MNPs) in the vicinity of HRSCs express similar regulatory mechanisms in parallel, and their presence in tissue is associated with inferior patient outcomes. MNPs in cHL have hitherto been identified by a small number of canonical markers and are usually described as tumor-associated macrophages. The organization of MNP networks and interactions with HRSCs remains unexplored at high resolution. Here, we defined the global immune-cell composition of cHL and nonlymphoma lymph nodes, integrating data across single-cell RNA sequencing, spatial transcriptomics, and multiplexed immunofluorescence. We observed that MNPs comprise multiple subsets of monocytes, macrophages, and dendritic cells (DCs). Classical monocytes, macrophages and conventional DC2s were enriched in the vicinity of HRSCs, but plasmacytoid DCs and activated DCs were excluded. Unexpectedly, cDCs and monocytes expressed immunoregulatory checkpoints PD-L1, TIM-3, and the tryptophan-catabolizing protein IDO, at the same level as macrophages. Expression of these molecules increased with age. We also found that classical monocytes are important signaling hubs, potentially controlling the retention of cDC2 and ThExh via CCR1-, CCR4-, CCR5-, and CXCR3-dependent signaling. Enrichment of the cDC2-monocyte-macrophage network in diagnostic biopsies is associated with early treatment failure. These results reveal unanticipated complexity and spatial polarization within the MNP compartment, further demonstrating their potential roles in immune evasion by cHL.

The impact of obesity and bariatric surgery on the immune microenvironment of the endometrium.

BACKGROUND: The incidence of endometrial cancer is rising in parallel with the obesity epidemic. Obesity increases endometrial cancer risk and weight loss is protective, but the underlying mechanisms are incompletely understood. We hypothesise that the immune microenvironment may influence susceptibility to malignant transformation in the endometrium. The aim of this study was to measure the impact of obesity and weight loss on the immunological landscape of the endometrium. METHODS: We conducted a prospective cohort study of women with class III obesity (body mass index, BMI ≥ 40 kg/m2) undergoing bariatric surgery or medically-supervised low-calorie diet. We collected blood and endometrial samples at baseline, and two and 12 months after weight loss intervention. Serum was analysed for inflammatory markers CRP, IL-6 and TNF-α. Multiplex immunofluorescence was used to simultaneously identify cells positive for immune markers CD68, CD56, CD3, CD8, FOXP3 and PD-1 in formalin-fixed paraffin-embedded endometrial tissue sections. Kruskal-Wallis tests were used to determine whether changes in inflammatory and immune biomarkers were associated with weight loss. RESULTS: Forty-three women with matched serum and tissue samples at all three time points were included in the analysis. Their median age and BMI were 44 years and 52 kg/m2, respectively. Weight loss at 12 months was greater in women who received bariatric surgery (n = 37, median 63.3 kg) than low-calorie diet (n = 6, median 12.8 kg). There were significant reductions in serum CRP (p = 3.62 × 10-6, r = 0.570) and IL-6 (p = 0.0003, r = 0.459), but not TNF-α levels, with weight loss. Tissue immune cell densities were unchanged except for CD8+ cells, which increased significantly with weight loss (p = 0.0097, r = -0.323). Tissue CD3+ cell density correlated negatively with systemic IL-6 levels (p = 0.0376; r = -0.318). CONCLUSION: Weight loss is associated with reduced systemic inflammation and a recruitment of protective immune cell types to the endometrium, supporting the concept that immune surveillance may play a role in endometrial cancer prevention.

Immune infiltrate diversity confers a good prognosis in follicular lymphoma.

BACKGROUND: Follicular lymphoma (FL) prognosis is influenced by the composition of the tumour microenvironment. We tested an automated approach to quantitatively assess the phenotypic and spatial immune infiltrate diversity as a prognostic biomarker for FL patients. METHODS: Diagnostic biopsies were collected from 127 FL patients initially treated with rituximab-based therapy (52%), radiotherapy (28%), or active surveillance (20%). Tissue microarrays were constructed and stained using multiplex immunofluorescence (CD4, CD8, FOXP3, CD21, PD-1, CD68, and DAPI). Subsequently, sections underwent automated cell scoring and analysis of spatial interactions, defined as cells co-occurring within 30 µm. Shannon's entropy, a metric describing species biodiversity in ecological habitats, was applied to quantify immune infiltrate diversity of cell types and spatial interactions. Immune infiltrate diversity indices were tested in multivariable Cox regression and Kaplan-Meier analysis for overall (OS) and progression-free survival (PFS). RESULTS: Increased diversity of cell types (HR = 0.19 95% CI 0.06-0.65, p = 0.008) and cell spatial interactions (HR = 0.39, 95% CI 0.20-0.75, p = 0.005) was associated with favourable OS, independent of the Follicular Lymphoma International Prognostic Index. In the rituximab-treated subset, the favourable trend between diversity and PFS did not reach statistical significance. CONCLUSION: Multiplex immunofluorescence and Shannon's entropy can objectively quantify immune infiltrate diversity and generate prognostic information in FL. This automated approach warrants validation in additional FL cohorts, and its applicability as a pre-treatment biomarker to identify high-risk patients should be further explored. The multiplex image dataset generated by this study is shared publicly to encourage further research on the FL microenvironment.

Spatial proximity between T and PD-L1 expressing cells as a prognostic biomarker for oropharyngeal squamous cell carcinoma.

BACKGROUND: Fulfilling the promise of cancer immunotherapy requires novel predictive biomarkers to characterise the host immune microenvironment. Deciphering the complexity of immune cell interactions requires an automated multiplex approach to histological analysis of tumour sections. We tested a new automatic approach to select tissue and quantify the frequencies of cell-cell spatial interactions occurring in the PD1/PD-L1 pathway, hypothesised to reflect immune escape in oropharyngeal squamous cell carcinoma (OPSCC). METHODS: Single sections of diagnostic biopsies from 72 OPSCC patients were stained using multiplex immunofluorescence (CD8, PD1, PD-L1, CD68). Following multispectral scanning and automated regions-of-interest selection, the Hypothesised Interaction Distribution (HID) method quantified spatial proximity between cells. Method applicability was tested by investigating the prognostic significance of co-localised cells (within 30 µm) in patients stratified by HPV status. RESULTS: High frequencies of proximal CD8+ and PD-L1+ (HR 2.95, p = 0.025) and PD1+ and PD-L1+ (HR 2.64, p = 0.042) cells were prognostic for poor overall survival in patients with HPV negative OPSCC (n = 31). CONCLUSION: The HID method can quantify spatial interactions considered to reflect immune escape and generate prognostic information in OPSCC. The new automated approach is ready to test in additional cohorts and its applicability should be explored in research and clinical studies.

Prediction of reader estimates of mammographic density using convolutional neural networks.

Mammographic density is an important risk factor for breast cancer. In recent research, percentage density assessed visually using visual analogue scales (VAS) showed stronger risk prediction than existing automated density measures, suggesting readers may recognize relevant image features not yet captured by hand-crafted algorithms. With deep learning, it may be possible to encapsulate this knowledge in an automatic method. We have built convolutional neural networks (CNN) to predict density VAS scores from full-field digital mammograms. The CNNs are trained using whole-image mammograms, each labeled with the average VAS score of two independent readers. Each CNN learns a mapping between mammographic appearance and VAS score so that at test time, they can predict VAS score for an unseen image. Networks were trained using 67,520 mammographic images from 16,968 women and for model selection we used a dataset of 73,128 images. Two case-control sets of contralateral mammograms of screen detected cancers and prior images of women with cancers detected subsequently, matched to controls on age, menopausal status, parity, HRT and BMI, were used for evaluating performance on breast cancer prediction. In the case-control sets, odd ratios of cancer in the highest versus lowest quintile of percentage density were 2.49 (95% CI: 1.59 to 3.96) for screen-detected cancers and 4.16 (2.53 to 6.82) for priors, with matched concordance indices of 0.587 (0.542 to 0.627) and 0.616 (0.578 to 0.655), respectively. There was no significant difference between reader VAS and predicted VAS for the prior test set (likelihood ratio chi square, p = 0.134 ). Our fully automated method shows promising results for cancer risk prediction and is comparable with human performance.