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.

Machine learning-based immune phenotypes correlate with STK11/KEAP1 co-mutations and prognosis in resectable NSCLC: a sub-study of the TNM-I trial

BACKGROUND

We aim to implement an immune cell score model in routine clinical practice for resected non-small-cell lung cancer (NSCLC) patients (NCT03299478). Molecular and genomic features associated with immune phenotypes in NSCLC have not been explored in detail.

PATIENTS AND METHODS

We developed a machine learning (ML)-based model to classify tumors into one of three categories: inflamed, altered, and desert, based on the spatial distribution of CD8+ T cells in two prospective (n = 453; TNM-I trial) and retrospective (n = 481) stage I-IIIA NSCLC surgical cohorts. NanoString assays and targeted gene panel sequencing were used to evaluate the association of gene expression and mutations with immune phenotypes.

RESULTS

Among the total of 934 patients, 24.4% of tumors were classified as inflamed, 51.3% as altered, and 24.3% as desert. There were significant associations between ML-derived immune phenotypes and adaptive immunity gene expression signatures. We identified a strong association of the nuclear factor-κB pathway and CD8+ T-cell exclusion through a positive enrichment in the desert phenotype. KEAP1 [odds ratio (OR) 0.27, Q = 0.02] and STK11 (OR 0.39, Q = 0.04) were significantly co-mutated in non-inflamed lung adenocarcinoma (LUAD) compared to the inflamed phenotype. In the retrospective cohort, the inflamed phenotype was an independent prognostic factor for prolonged disease-specific survival and time to recurrence (hazard ratio 0.61, P = 0.01 and 0.65, P = 0.02, respectively).

CONCLUSIONS

ML-based immune phenotyping by spatial distribution of T cells in resected NSCLC is able to identify patients at greater risk of disease recurrence after surgical resection. LUADs with concurrent KEAP1 and STK11 mutations are enriched for altered and desert immune phenotypes.