Cancer Subtype Classification Research Articles

REMR: Identification of RNA Editing-mediated MiRNA Regulation in Cancers

Dysregulation of adenosine-to-inosine (A-to-I) RNA editing has been implicated in cancer progression. However, a comprehensive understanding of how A-to-I RNA editing is incorporated into miRNA regulation to modulate gene expression in cancer remains unclear, given the lack of effective identification methods. To this end, we introduced an information theory-based algorithm named REMR to systematically identify 12,006 A-to-I RNA editing-mediated miRNA regulatory triplets (RNA editing sites, miRNAs, and genes) across ten major cancer types based on multi-omics profiling data from The Cancer Genome Atlas (TCGA). Through analyses of functional enrichment, transcriptional regulatory networks, and protein-protein interaction (PPI) networks, we showed that RNA editing-mediated miRNA regulation potentially affects critical cancer-related functions, such as apoptosis, cell cycle, drug resistance, and immunity. Furthermore, triplets can serve as biomarkers for classifying cancer subtypes with distinct prognoses or drug responses, highlighting the clinical relevance of such regulation. In addition, an online resource (http://www.jianglab.cn/REMR/) was constructed to support the convenient retrieval of our findings. In summary, our study systematically dissected the RNA editing-mediated miRNA regulations, thereby providing a valuable resource for understanding the mechanism of RNA editing as an epitranscriptomic regulator in cancer.

MoAGL-SA: a multi-omics adaptive integration method with graph learning and self attention for cancer subtype classification

BackgroundThe integration of multi-omics data through deep learning has greatly improved cancer subtype classification, particularly in feature learning and multi-omics data integration. However, key challenges remain in embedding sample structure information into the feature space and designing flexible integration strategies.ResultsWe propose MoAGL-SA, an adaptive multi-omics integration method based on graph learning and self-attention, to address these challenges. First, patient relationship graphs are generated from each omics dataset using graph learning. Next, three-layer graph convolutional networks are employed to extract omic-specific graph embeddings. Self-attention is then used to focus on the most relevant omics, adaptively assigning weights to different graph embeddings for multi-omics integration. Finally, cancer subtypes are classified using a softmax classifier.ConclusionsExperimental results show that MoAGL-SA outperforms several popular algorithms on datasets for breast invasive carcinoma, kidney renal papillary cell carcinoma, and kidney renal clear cell carcinoma. Additionally, MoAGL-SA successfully identifies key biomarkers for breast invasive carcinoma.

Abstract A004: Full spectrum isolation technique-based dual circulating rare cells analysis to maximize early cancer diagnosis

Abstract Liquid biopsy (LB) is essential in precision medicine, enabling real-time tumor monitoring through repeated sampling. However, its diagnostic sensitivity and accuracy are limited by the low detection rate of circulating tumor cells (CTCs). For example, in breast cancer, the CTC detection rate exhibited only 20%, significantly limiting its clinical utility. This limitation arises since existing technologies selectively isolate specific phenotypes of CTCs, failing to capture others, such as mesenchymal CTCs, which are associated with poor prognosis. These mesenchymal CTCs often evade detection due to their lack of EpCAM expression and high deformability, making difficult to isolate them with traditional filter or EpCAM based methods. Additionally, conventional LB methods fail to provide critical insights into the tumor microenvironment (TME). To overcome these limitations, we developed a novel approach that improves LB diagnostic sensitivity and accuracy by (1) creating a technology to isolate all types of CTCs, regardless of marker expression or size, and (2) simultaneously analyzing CTCs and circulating cancer-associated fibroblasts (cCAFs). This dual approach addresses the challenges of low sensitivity and accuracy by enabling the isolation of a broader range of CTCs, including mesenchymal types, and by incorporating cCAF analysis. cCAFs, derived from the TME and found in the bloodstream, provide valuable insights into the TME that are hard to obtain through traditional methods. Analyzing cCAFs alongside CTCs offers a more comprehensive understanding of tumor biology and its microenvironment, leading to improved diagnostic accuracy, especially in early-stage cancer. In our study involving 55 breast cancer patients, we demonstrated that while CTCs were undetectable in 42.9% of early-stage patients, cCAFs were detected in all cases and were found to be 10 times more abundant than CTCs. Importantly, for the first time, this study uncovers that the diverse CAF subtypes previously identified in tissue samples also exist in the bloodstream, highlighting their critical role in providing TME-related insights. In addition, cCAF subtypes vary with hormonal receptor expression, and their marker analysis allows for 80.9% accurate classification of breast cancer subtypes. We applied this approach to early-stage breast cancer patients, for whom diagnosis through liquid biopsy is particularly difficult, and found that the combined analysis of CTCs and cCAFs achieved a 18% improvement in accuracy, with the AUC increasing from 0.6817 to 0.8009 compared to CTC analysis alone. This innovative approach significantly enhances LB by simultaneously isolating all CTC subtypes and incorporating cCAF analysis, thereby offering a more sensitive and accurate diagnostic tool. Beyond improving breast cancer diagnostics, this method has the potential to be applied to other tumor types and could also prove valuable for prognosis prediction and therapeutic response monitoring, ultimately leading to better patient outcomes and more effective cancer treatment strategies. Citation Format: Hyeongjung Woo, Simon A Joosse, Frederike Harten, Felix Hilpert, Klaus Pantel, Minseok S. Kim. Full spectrum isolation technique-based dual circulating rare cells analysis to maximize early cancer diagnosis [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr A004.

LW-MorphCNN: a lightweight morphological attention-based subtype classification network for lung cancer

Abstract Lung cancer is generally considered one of the most deadly cancers globally. If it can be identified early and diagnosed correctly, the survival probability of patients can be significantly improved. In this process, histopathological examination is a commonly used method for diagnosing and detecting lung cancer. It is crucial to accurately identify lung cancer subtypes from histopathological images, as this helps doctors formulate effective treatment plans. However, the visual inspection in histopathological diagnosis requires a large amount of time and also depends on the subjective perception of clinicians. Therefore, this paper proposes a lightweight lung cancer subtype classification network based on morphological attention (LW-MorphCNN), which is used to automatically classify the histopathological images of benign lung tumors, ADC (adenocarcinoma), and SCC (squamous cell carcinoma) provided in the public dataset LC25000 (Lung and Colon). This paper takes histopathological images as input and conducts a comparative analysis with classic networks such as VGG16, VGG19, DenseNet121, and ResNet50, as well as existing classification methods proposed in the same work. The network proposed in this paper is superior to other networks in terms of parameter quantity and performance, with an accuracy rate and F1 - score reaching 99.47% and 99.44% respectively. Clinicians can install the provided LW-MorphCNN in the hospital to confirm the diagnosis results.

Multimodal Co-Attention Fusion Network With Online Data Augmentation for Cancer Subtype Classification.

It is an essential task to accurately diagnose cancer subtypes in computational pathology for personalized cancer treatment. Recent studies have indicated that the combination of multimodal data, such as whole slide images (WSIs) and multi-omics data, could achieve more accurate diagnosis. However, robust cancer diagnosis remains challenging due to the heterogeneity among multimodal data, as well as the performance degradation caused by insufficient multimodal patient data. In this work, we propose a novel multimodal co-attention fusion network (MCFN) with online data augmentation (ODA) for cancer subtype classification. Specifically, a multimodal mutual-guided co-attention (MMC) module is proposed to effectively perform dense multimodal interactions. It enables multimodal data to mutually guide and calibrate each other during the integration process to alleviate inter- and intra-modal heterogeneities. Subsequently, a self-normalizing network (SNN)-Mixer is developed to allow information communication among different omics data and alleviate the high-dimensional small-sample size problem in multi-omics data. Most importantly, to compensate for insufficient multimodal samples for model training, we propose an ODA module in MCFN. The ODA module leverages the multimodal knowledge to guide the data augmentations of WSIs and maximize the data diversity during model training. Extensive experiments are conducted on the public TCGA dataset. The experimental results demonstrate that the proposed MCFN outperforms all the compared algorithms, suggesting its effectiveness.

Ensemble transformer-based multiple instance learning to predict pathological subtypes and tumor mutational burden from histopathological whole slide images of endometrial and colorectal cancer

In endometrial cancer (EC) and colorectal cancer (CRC), in addition to microsatellite instability, tumor mutational burden (TMB) has gradually gained attention as a genomic biomarker that can be used clinically to determine which patients may benefit from immune checkpoint inhibitors. High TMB is characterized by a large number of mutated genes, which encode aberrant tumor neoantigens, and implies a better response to immunotherapy. Hence, a part of EC and CRC patients associated with high TMB may have higher chances to receive immunotherapy. TMB measurement was mainly evaluated by whole-exome sequencing or next-generation sequencing, which was costly and difficult to be widely applied in all clinical cases. Therefore, an effective, efficient, low-cost and easily accessible tool is urgently needed to distinguish the TMB status of EC and CRC patients. In this study, we present a deep learning framework, namely Ensemble Transformer-based Multiple Instance Learning with Self-Supervised Learning Vision Transformer feature encoder (ETMIL-SSLViT), to predict pathological subtype and TMB status directly from the H&E stained whole slide images (WSIs) in EC and CRC patients, which is helpful for both pathological classification and cancer treatment planning. Our framework was evaluated on two different cancer cohorts, including an EC cohort with 918 histopathology WSIs from 529 patients and a CRC cohort with 1495 WSIs from 594 patients from The Cancer Genome Atlas. The experimental results show that the proposed methods achieved excellent performance and outperforming seven state-of-the-art (SOTA) methods in cancer subtype classification and TMB prediction on both cancer datasets. Fisher’s exact test further validated that the associations between the predictions of the proposed models and the actual cancer subtype or TMB status are both extremely strong (p<0.001). These promising findings show the potential of our proposed methods to guide personalized treatment decisions by accurately predicting the EC and CRC subtype and the TMB status for effective immunotherapy planning for EC and CRC patients.

Designing Optimal Probe Sequence for Breast Cancer Subtype Classification via Multiple MicroRNAs Recognition

Designing Optimal Probe Sequence for Breast Cancer Subtype Classification via Multiple MicroRNAs Recognition

Navigating heme pathways: the breach of heme oxygenase and hemin in breast cancer.

Breast cancer remains a significant global health challenge, with diverse subtypes and complex molecular mechanisms underlying its development and progression. This review comprehensively examines recent advances in breast cancer research, with a focus on classification, molecular pathways, and the role of heme oxygenases (HO), heme metabolism implications, and therapeutic innovations. The classification of breast cancer subtypes based on molecular profiling has significantly improved diagnosis and treatment strategies, allowing for tailored approaches to patient care. Molecular studies have elucidated key signaling pathways and biomarkers implicated in breast cancer pathogenesis, shedding light on potential targets for therapeutic intervention. Notably, emerging evidence suggests a critical role for heme oxygenases, particularly HO-1, in breast cancer progression and therapeutic resistance, highlighting the importance of understanding heme metabolism in cancer biology. Furthermore, this review highlights recent advances in breast cancer therapy, including targeted therapies, immunotherapy, and novel drug delivery systems. Understanding the complex interplay between breast cancer subtypes, molecular pathways, and innovative therapeutic approaches is essential for improving patient outcomes and developing more effective treatment strategies in the fight against breast cancer.

Decoding Breast Cancer: Using Radiomics to Non-Invasively Unveil Molecular Subtypes Directly from Mammographic Images.

Breast cancer is the most commonly diagnosed cancer worldwide. The therapy used and its success depend highly on the histology of the tumor. This study aimed to explore the potential of predicting the molecular subtype of breast cancer using radiomic features extracted from screening digital mammography (DM) images. A retrospective study was performed using the OPTIMAM Mammography Image Database (OMI-DB). Four binary classification tasks were performed: luminal A vs. non-luminal A, luminal B vs. non-luminal B, TNBC vs. non-TNBC, and HER2 vs. non-HER2. Feature selection was carried out by Pearson correlation and LASSO. The support vector machine (SVM) and naive Bayes (NB) ML classifiers were used, and their performance was evaluated with the accuracy and the area under the receiver operating characteristic curve (AUC). A total of 186 patients were included in the study: 58 luminal A, 35 luminal B, 52 TNBC, and 41 HER2. The SVM classifier resulted in AUCs during testing of 0.855 for luminal A, 0.812 for luminal B, 0.789 for TNBC, and 0.755 for HER2, respectively. The NB classifier showed AUCs during testing of 0.714 for luminal A, 0.746 for luminal B, 0.593 for TNBC, and 0.714 for HER2. The SVM classifier outperformed NB with statistical significance for luminal A (p = 0.0268) and TNBC (p = 0.0073). Our study showed the potential of radiomics for non-invasive breast cancer subtype classification.

MBFusion: Multi-modal balanced fusion and multi-task learning for cancer diagnosis and prognosis

Pathological images and molecular omics are important information for predicting diagnosis and prognosis. The two kinds of heterogeneous modal data contain complementary information, and the effective fusion of the two modals can better reveal the complex mechanisms of cancer. However, due to the different representation learning methods, the expression strength of different modals in different tasks varies greatly, so that many multimodal fusions do not achieve the best results. In this paper, MBFusion is proposed, to achieve multiple tasks such as prediction of diagnosis and prognosis through multi-modal balanced fusion. The MBFusion framework uses two kinds of specially constructed graph convolutional network to extract the features of molecular omics data, and uses ResNet to extract the features of pathological image data and retain important deep features by using attention and clustering, which effectively improves both kinds of the features representation, making their expressive ability balanced and comparable. The features of these two modal data are then fused through cross-attention Transformer, and the fused features are used to learn both tasks of cancer subtype classification and survival analysis by using multi-task learning. In this paper, MBFusion and other state of the art methods are compared on two public cancer datasets, and MBFusion shows an improvement of up to 10.1% by three kinds of evaluation metrics. In the ablation experiment, MBFusion explores the contribution of each modal data and each framework module to the performance. Furthermore, the interpretability of MBFusion is explained in detail to show the value of application.

Breast Cancer Classification From Digital Pathology Images via Connectivity-Aware Graph Transformer.

Automated classification of breast cancer subtypes from digital pathology images has been an extremely challenging task due to the complicated spatial patterns of cells in the tissue micro-environment. While newly proposed graph transformers are able to capture more long-range dependencies to enhance accuracy, they largely ignore the topological connectivity between graph nodes, which is nevertheless critical to extract more representative features to address this difficult task. In this paper, we propose a novel connectivity-aware graph transformer (CGT) for phenotyping the topology connectivity of the tissue graph constructed from digital pathology images for breast cancer classification. Our CGT seamlessly integrates connectivity embedding to node feature at every graph transformer layer by using local connectivity aggregation, in order to yield more comprehensive graph representations to distinguish different breast cancer subtypes. In light of the realistic intercellular communication mode, we then encode the spatial distance between two arbitrary nodes as connectivity bias in self-attention calculation, thereby allowing the CGT to distinctively harness the connectivity embedding based on the distance of two nodes. We extensively evaluate the proposed CGT on a large cohort of breast carcinoma digital pathology images stained by Haematoxylin & Eosin. Experimental results demonstrate the effectiveness of our CGT, which outperforms state-of-the-art methods by a large margin. Codes are released on https://github.com/wang-kang-6/CGT.

Deep Learning-Based Ovarian Cancer Subtype Classification Using VGG16 and MobileNetV2 with Squeeze-and-Excitation Blocks

Background: Ovarian carcinoma remains one of the deadliest gynecological malignancies due to its heterogeneity and late-stage detection. Accurate classification of its subtypes—High-Grade Serous Carcinoma (HGSC), Clear-Cell Carcinoma (CC), Endometrioid Carcinoma (EC), Low-Grade Serous Carcinoma (LGSC), and Mucinous Carcinoma (MC)—is essential for tailored treatments, yet traditional histopathological methods often lack precision. This study aimed to develop a deep learning (DL) model to enhance ovarian cancer subtype classification using histopathological images. Methods: Histopathological images were collected from medical repositories, and data augmentation techniques were applied to increase dataset diversity. Two convolutional neural network (CNN) architectures, VGG16 and MobileNetV2, were fine-tuned using transfer learning to classify the subtypes. The models were pre-trained on the ImageNet dataset and evaluated through accuracy, precision, recall, F1-score, and ROC-AUC, with K-fold cross-validation ensuring robustness. Results: Results indicated that VGG16 improved classification over baseline CNN models, while MobileNetV2, with Squeeze-and-Excitation (SE) blocks, achieved the highest performance, offering greater accuracy and computational efficiency. MobileNetV2’s lightweight architecture captured intricate tissue patterns more effectively, making it the superior model. Conclusion: This study highlights the potential of advanced DL models, particularly MobileNetV2 with SE block attention, for improving ovarian cancer subtype classification. These findings offer promising implications for clinical practice and personalized treatment approaches. Future research should focus on larger datasets and integrating multimodal data for further advancements.

Deep fine-KNN classification of ovarian cancer subtypes using efficientNet-B0 extracted features: a comprehensive analysis

This study presents a robust approach for the classification of ovarian cancer subtypes through the integration of deep learning and k-nearest neighbor (KNN) methods. The proposed model leverages the powerful feature extraction capabilities of EfficientNet-B0, utilizing its deep features for subsequent fine-grained classification using the fine-KNN approach. The UBC-OCEAN dataset, encompassing histopathological images of five distinct ovarian cancer subtypes, namely, high-grade serous carcinoma (HGSC), clear-cell ovarian carcinoma (CC), endometrioid carcinoma (EC), low-grade serous carcinoma (LGSC), and mucinous carcinoma (MC), served as the foundation for our investigation. With a dataset comprising 725 images, divided into 80% for training and 20% for testing, our model exhibits exceptional performance. Both the validation and testing phases achieved 100% accuracy, underscoring the efficacy of the proposed methodology. In addition, the area under the curve (AUC), a key metric for evaluating the model’s discriminative ability, demonstrated high performance across various subtypes, with AUC values of 0.94, 0.78, 0.69, 0.92, and 0.94 for MC. Furthermore, the positive likelihood ratios (LR+) were indicative of the model’s diagnostic utility, with notable values for each subtype: CC (27.294), EC (9.441), HGSC (12.588), LGSC (17.942), and MC (17.942). These findings demonstrate the effectiveness of the model in distinguishing between ovarian cancer subtypes, positioning it as a promising tool for diagnostic applications. The demonstrated accuracy, AUC values, and LR+ values underscore the potential of the model as a valuable diagnostic tool, contributing to the advancement of precision medicine in the field of ovarian cancer research.

Noninvasive Lung Cancer Subtype Classification Using Tumor-Derived Signatures and cfDNA Methylome.

This study explores the use of cfDNA methylomes for the classification of lung cancer subtypes, vital for effective treatment. By identifying specific methylation markers in tumor tissues, we developed a robust classification model achieving high accuracy for noninvasive subtype detection. This cfDNA methylome approach offers promising avenues for early detection and monitoring.

Multinomial Logistic Regression with Adaptive Regularization for Cancer Subtype Classification via Multi-omics Data

Background: Integrating multi-omics data for cancer classification brings complementary biological insights while also facing challenges such as data integration, gene grouping, and adaptive weight construction. Objective: This paper aims to address the challenges faced by the cancer subtype classification and gene screening based on multi-omics data. Methods: Multinomial logistic regression with adaptive regularization (MLRAR) was proposed by integrating DNA methylation, gene mutation, and RNA-seq information. A data preprocessing strategy that effectively utilizes multi-omics information was presented, and the local maximum quasiclique merging (lmQCM) algorithm was implemented to group genes. Biological pathway information was utilized to evaluate the significance of gene groups, while the significance of each gene within a group was evaluated by integrating mutation information, information theory, and methylation information. Results: Compared to MRlasso, MRGL, MSGL, MROGL, AMRSOGL, and AGLRMR, the proposed method yielded improvements in subtype classification accuracy of breast cancer by 2.6%, 2.9%, 3.5%, 2.3%, 2.0%, and 1.8%, respectively. In addition, MLRAR also achieved significant improvements in ovarian cancer by 8.2%, 5.0%, 6.8%, 5.2%, 12.7%, and 6.3%, respectively. Conclusion: The proposed method can effectively deal with data integration, gene grouping, and adaptive weight construction.

NNBGWO-BRCA marker: Neural Network and binary grey wolf optimization based Breast cancer biomarker discovery framework using multi-omics dataset

Background and ObjectiveBreast cancer is a multifaceted condition characterized by diverse features and a substantial mortality rate, underscoring the imperative for timely detection and intervention. The utilization of multi-omics data has gained significant traction in recent years to identify biomarkers and classify subtypes in breast cancer. This kind of research idea from part to whole will also be an inevitable trend in future life science research. Deep learning can integrate and analyze multi-omics data to predict cancer subtypes, which can further drive targeted therapies. However, there are few articles leveraging the nature of deep learning for feature selection. Therefore, this paper proposes a Neural Network and Binary grey Wolf Optimization based BReast CAncer bioMarker (NNBGWO-BRCAMarker) discovery framework using multi-omics data to obtain a series of biomarkers for precise classification of breast cancer subtypes. MethodsNNBGWO-BRCAMarker consists of two phases: in the first phase, relevant genes are selected using the weights obtained from a trained feedforward neural network; in the second phase, the binary grey wolf optimization algorithm is leveraged to further screen the selected genes, resulting in a set of potential breast cancer biomarkers. ResultsThe SVM classifier with RBF kernel achieved a classification accuracy of 0.9242 ± 0.03 when trained using the 80 biomarkers identified by NNBGWO-BRCAMarker, as evidenced by the experimental results. We conducted a comprehensive gene set analysis, prognostic analysis, and druggability analysis, unveiling 25 druggable genes, 16 enriched pathways strongly linked to specific subtypes of breast cancer, and 8 genes linked to prognostic outcomes. ConclusionsThe proposed framework successfully identified 80 biomarkers from the multi-omics data, enabling accurate classification of breast cancer subtypes. This discovery may offer novel insights for clinicians to pursue in further studies.

Development and validation of mFISHseq: A diagnostic test using multiplexed RNA-FISH–guided laser capture microdissection RNA sequencing.

3069 Background: Breast cancer (BCa) is a heterogeneous disease requiring precise diagnostic tools to guide effective treatment strategies. Current diagnostic assays, including various multigene assays, often fail to adequately address the complex biology of BCa subtypes. To address these limitations and enhance the understanding of BCa biology, we developed and validated a novel diagnostic, prognostic, and predictive tool, called mFISHseq. Methods: Our approach, mFISHseq, integrates multiplexed fluorescent in situ hybridization (FISH) of the four main BCa biomarkers, estrogen ( ESR1)/progesterone ( PGR)/Her2 ( ERBB2) receptors and Ki67 ( MKI67), which are used to guide laser capture microdissection (LCM) of regions of interest followed by RNA-sequencing. This technique ensures tumor purity, facilitates interrogation of tumor heterogeneity, consequently permitting unbiased analysis of whole transcriptome profiling data and explicitly quantifying the variability between different tumor regions. We validated mFISHseq on a retrospective cohort study involving 1,082 FFPE breast tumors with detailed clinicopathological data, informed consent, and ethical committee approval. Results: mFISHseq demonstrated excellent analytical validity with a 93% accuracy rate compared to standard immunohistochemistry (IHC), while providing more quantitative biomarker expression. Prespecified threshold values for mFISHseq derived from a split 70:30 training/test set showed exceptional concordance with IHC as demonstrated by area under the receiver operating characteristic (ROC) curves for all markers (AUC: MKI67=0.98, ERBB2=0.95, ESR1=0.95, PGR=0.93). Both RNA-FISH and -SEQ showed moderate to very strong correlations (Spearman’s r; ERBB2=0.41, MKI67=0.61, PGR=0.66, ESR1=0.75), thus highlighting the potential to use both orthogonal methodologies to cross-validate results. To demonstrate clinical validity, we developed a 293-gene intrinsic subtype classifier, showing substantial agreement to established classifiers like PAM50 and AIMS (Cohen’s κ= 0.75 and 0.73, respectively) and superior prognostic performance. We also report that LCM is an essential component of the mFISHseq workflow, since samples that did not undergo LCM showed reduced biomarker expression, elevated non-tumor gene expression, and misclassification of samples into less aggressive molecular subtypes (e.g., normal-like) and prognostic risk groups (e.g., high to low). Conclusions: The mFISHseq method showed excellent concordance with IHC and the use of LCM provides tumor-enriched samples that are devoid of contamination from non-tumor elements, thus providing unbiased spatially resolved interrogation of tumor heterogeneity. Altogether, mFISHseq solves a long-standing challenge in the precise diagnosis and classification of breast cancer subtypes.

Masked pre-training of transformers for histology image analysis

In digital pathology, whole-slide images (WSIs) are widely used for applications such as cancer diagnosis and prognosis prediction. Vision transformer (ViT) models have recently emerged as a promising method for encoding large regions of WSIs while preserving spatial relationships among patches. However, due to the large number of model parameters and limited labeled data, applying transformer models to WSIs remains challenging. In this study, we propose a pretext task to train the transformer model in a self-supervised manner. Our model, MaskHIT, uses the transformer output to reconstruct masked patches, measured by contrastive loss. We pre-trained MaskHIT model using over 7000 WSIs from TCGA and extensively evaluated its performance in multiple experiments, covering survival prediction, cancer subtype classification, and grade prediction tasks. Our experiments demonstrate that the pre-training procedure enables context-aware understanding of WSIs, facilitates the learning of representative histological features based on patch positions and visual patterns, and is essential for the ViT model to achieve optimal results on WSI-level tasks. The pre-trained MaskHIT surpasses various multiple instance learning approaches by 3% and 2% on survival prediction and cancer subtype classification tasks, and also outperforms recent state-of-the-art transformer-based methods. Finally, a comparison between the attention maps generated by the MaskHIT model with pathologist's annotations indicates that the model can accurately identify clinically relevant histological structures on the whole slide for each task.

Unsupervised mutual transformer learning for multi-gigapixel Whole Slide Image classification

The classification of gigapixel Whole Slide Images (WSIs) is an important task in the emerging area of computational pathology. There has been a surge of interest in deep learning models for WSI classification with clinical applications such as cancer detection or prediction of cellular mutations. Most supervised methods require expensive and labor-intensive manual annotations by expert pathologists. Weakly supervised Multiple Instance Learning (MIL) methods have recently demonstrated excellent performance; however, they still require large-scale slide-level labeled training datasets that require a careful inspection of each slide by an expert pathologist. In this work, we propose a fully unsupervised WSI classification algorithm based on mutual transformer learning. The instances (i.e., patches) from gigapixel WSIs are transformed into a latent space and then inverse-transformed to the original space. Using the transformation loss, pseudo labels are generated and cleaned using a transformer label cleaner. The proposed transformer-based pseudo-label generator and cleaner modules mutually train each other iteratively in an unsupervised manner. A discriminative learning mechanism is introduced to improve normal versus cancerous instance labeling. In addition to the unsupervised learning, we demonstrate the effectiveness of the proposed framework for weakly supervised learning and cancer subtype classification as downstream analysis. Extensive experiments on four publicly available datasets show better performance of the proposed algorithm compared to the existing state-of-the-art methods.

Luminal androgen receptor subtype and tumor-infiltrating lymphocytes groups based on triple-negative breast cancer molecular subclassification.

In our previous study, we developed a triple-negative breast cancer (TNBC) subtype classification that correlated with the TNBC molecular subclassification. In this study, we aimed to evaluate the predictor variables of this subtype classification on the whole slide and to validate the model's performance by using an external test set. We explored the characteristics of this subtype classification and investigated genomic alterations, including genomic scar signature scores. First, TNBC was classified into the luminal androgen receptor (LAR) and non-luminal androgen receptor (non-LAR) subtypes based on the AR Allred score (≥ 6 and < 6, respectively). Then, the non-LAR subtype was further classified into the lymphocyte-predominant (LP), lymphocyte-intermediate (LI), and lymphocyte-depleted (LD) groups based on stromal tumor-infiltrating lymphocytes (TILs) (< 20%, > 20% but < 60%, and ≥ 60%, respectively). This classification showed fair agreement with the molecular classification in the test set. The LAR subtype was characterized by a high rate of PIK3CA mutation, CD274 (encodes PD-L1) and PDCD1LG2 (encodes PD-L2) deletion, and a low homologous recombination deficiency (HRD) score. The non-LAR LD TIL group was characterized by a high frequency of NOTCH2 and MYC amplification and a high HRD score.