Predicting Breast Cancer Gene Expression Signature by Applying Deep Convolutional Neural Networks From Unannotated Pathological Images

Development and validation of a predictive model combining radiomics and deep learning features for spread through air spaces in stage T1 non-small cell lung cancer: a multicenter study

Purpose

The goal of this paper is to compare the effectiveness of three deep learning models (2D, 3D, and 2.5D), three radiomics models(INTRA, Peri2mm, and Fusion2mm), and a combined model in predicting the spread through air spaces (STAS) in non-small cell lung cancer (NSCLC) to identify the optimal model for clinical surgery planning.

Methods

We included 480 patients who underwent surgery at four centers between January 2019 and August 2024, dividing them into a training cohort, an internal test cohort, and an external validation cohort. We extracted deep learning features using the ResNet50 algorithm. Least absolute shrinkage selection operator(Lasso) and spearman rank correlation were utilized to choose features. Extreme Gradient Boosting (XGboost) was used to execute deep learning and radiomics. Then, a combination model was developed, integrating both sources of data.

Result

The combined model showed outstanding performance, with an area under the receiver operating characteristic curve (AUC) of 0.927 (95% CI 0.870 - 0.984) in the test set and 0.867 (95% CI 0.819 - 0.915) in the validation set. This model significantly distinguished between high-risk and low-risk patients and demonstrated significant advantages in clinical application.

Conclusion

The combined model is adequate for preoperative prediction of STAS in patients with stage T1 NSCLC, outperforming the other six models in predicting STAS risk.

Image analysis-based identification of high risk ER-positive, HER2-negative breast cancers

BACKGROUND

Breast cancer subtypes Luminal A and Luminal B are classified by the expression of PAM50 genes and may benefit from different treatment strategies. Machine learning models based on H&E images may contain features associated with subtype, allowing early identification of tumors with higher risk of recurrence.

METHODS

H&E images (n = 630 ER+/HER2-breast cancers) were pixel-level segmented into epithelium and stroma. Convolutional neural network and multiple instance learning were used to extract image features from original and segmented images. Patient-level classification models were trained to discriminate Luminal A versus B image features in tenfold cross-validation, with or without grade adjustment. The best-performing visual classifier was incorporated into envisioned diagnostic protocols as an alternative to genomic testing (PAM50). The protocols were then compared in time-to-recurrence models.

RESULTS

Among ER+/HER2-tumors, the image-based protocol differentiated recurrence times with a hazard ratio (HR) of 2.81 (95% CI: 1.73-4.56), which was similar to the HR for PAM50 (2.66, 95% CI: 1.65-4.28). Grade adjustment did not improve subtype prediction accuracy, but did help balance sensitivity and specificity. Among high grade participants, sensitivity and specificity (0.734 and 0.474, respectively) became more similar (0.732 and 0.624, respectively) in grade-adjusted models. The original and epithelium-specific images had similar performance and highest accuracy, followed by stroma or binarized images showing only the epithelial-stromal interface.

CONCLUSIONS

Given low rates of genomic testing uptake nationally, image-based methods may help identify ER+/HER2-patients who could benefit from testing.

Publicly available datasets of breast histopathology H&E whole-slide images: A scoping review

Advancements in digital pathology and computing resources have made a significant impact in the field of computational pathology for breast cancer diagnosis and treatment. However, access to high-quality labeled histopathological images of breast cancer is a big challenge that limits the development of accurate and robust deep learning models. In this scoping review, we identified the publicly available datasets of breast H&E-stained whole-slide images (WSIs) that can be used to develop deep learning algorithms. We systematically searched 9 scientific literature databases and 9 research data repositories and found 17 publicly available datasets containing 10 385 H&E WSIs of breast cancer. Moreover, we reported image metadata and characteristics for each dataset to assist researchers in selecting proper datasets for specific tasks in breast cancer computational pathology. In addition, we compiled 2 lists of breast H&E patches and private datasets as supplementary resources for researchers. Notably, only 28% of the included articles utilized multiple datasets, and only 14% used an external validation set, suggesting that the performance of other developed models may be susceptible to overestimation. The TCGA-BRCA was used in 52% of the selected studies. This dataset has a considerable selection bias that can impact the robustness and generalizability of the trained algorithms. There is also a lack of consistent metadata reporting of breast WSI datasets that can be an issue in developing accurate deep learning models, indicating the necessity of establishing explicit guidelines for documenting breast WSI dataset characteristics and metadata.

Artificial intelligence: illuminating the depths of the tumor microenvironment

Artificial intelligence (AI) can acquire characteristics that are not yet known to humans through extensive learning, enabling to handle large amounts of pathology image data. Divided into machine learning and deep learning, AI has the advantage of handling large amounts of data and processing image analysis, consequently it also has a great potential in accurately assessing tumour microenvironment (TME) models. With the complex composition of the TME, in-depth study of TME contributes to new ideas for treatment, assessment of patient response to postoperative therapy and prognostic prediction. This leads to a review of the development of AI's application in TME assessment in this study, provides an overview of AI techniques applied to medicine, delves into the application of AI in analysing the quantitative and spatial location characteristics of various cells (tumour cells, immune and non-immune cells) in the TME, reveals the predictive prognostic value of TME and provides new ideas for tumour therapy, highlights the great potential for clinical applications. In addition, a discussion of its limitations and encouraging future directions for its practical clinical application is presented.

iCluF: an unsupervised iterative cluster-fusion method for patient stratification using multiomics data

Motivation

Patient stratification is crucial for the effective treatment or management of heterogeneous diseases, including cancers. Multiomic technologies facilitate molecular characterization of human diseases; however, the complexity of data warrants the need for the development of robust data integration tools for patient stratification using machine-learning approaches.

Results

iCluF iteratively integrates three types of multiomic data (mRNA, miRNA, and DNA methylation) using pairwise patient similarity matrices built from each omic data. The intermediate omic-specific neighborhood matrices implement iterative matrix fusion and message passing among the similarity matrices to derive a final integrated matrix representing all the omics profiles of a patient, which is used to further cluster patients into subtypes. iCluF outperforms other methods with significant differences in the survival profiles of 8581 patients belonging to 30 different cancers in TCGA. iCluF also predicted the four intrinsic subtypes of Breast Invasive Carcinomas with adjusted rand index and Fowlkes-Mallows scores of 0.72 and 0.83, respectively. The Gini importance score showed that methylation features were the primary decisive players, followed by mRNA and miRNA to identify disease subtypes. iCluF can be applied to stratify patients with any disease containing multiomic datasets.

Availability and implementation

Source code and datasets are available at https://github.com/GudaLab/iCluF_core.