AI identifies new high-risk subtypes of endometrial cancer

In a recently published study, Nature CommunicationsA team of researchers used artificial intelligence (AI) to classify histopathological images to distinguish between endometrial cancer subtypes. The tool identified a subtype of endometrial cancer called No Specific Molecular Profile (NSMP), which is characterized by aggressive disease and poor survival.

background

Endometrial cancer is classified into four subtypes, each with significantly different therapeutic implications and prognosis.

Until now, classification of these subtypes has been based on poorly reproducible clinicopathological parameters, which has had direct implications on cancer management.

Inconsistencies in the assignment of tumor histology and grade can lead to inaccurate risk assessment, overtreatment or inappropriate treatment, which can result in recurrence and death.

Using exome and whole genome sequencing and microsatellite instability assays, The Cancer Genome Atlas Project has shown that endometrial cancer can be classified into four prognostic subtypes based on major genetic alterations.

Furthermore, the development of AI tools using deep learning models is increasingly being applied in medicine to process large amounts of image and text data, which is then used to identify potential biomarkers and improve cancer pathology diagnosis.

About the Research

In this study, the researchers built an AI-based image classification tool that used deep learning capabilities to analyze histopathological images of hematoxylin and eosin-stained slides to distinguish between two subtypes of endometrial cancer: NSMP and p53 abnormality or p53abn.

In a previous study, researchers developed a molecular classification system for endometrial cancer that can be easily applied in clinical settings, dividing endometrial cancer into four subtypes:

Initially Pole The mutational subtype contained pathogenic mutations in a gene (DNA polymerase epsilon or POLE) involved in deoxyribonucleic acid (DNA) proofreading and repair.

The second subtype is the mismatch repair deficient subtype (MMRd), where immunohistochemistry-based diagnostic tests reveal the absence of key proteins involved in mismatch repair.

The third subtype was also diagnosed using immunohistochemical analysis and was characterized by abnormalities in the p53 tumor suppressor protein.

The last subtype, NSMP, was diagnosed by excluding all diagnostic characteristics of the other three subtypes, as it has no defining features.

Here, the researchers used AI-based image classification to analyze histopathological features and differentiate between subtypes NSMP and p53abn.

They hypothesized that some patients within the NSMP subtype have tumors that are histologically similar to those seen in patients within the p53abn subtype, and that application of deep learning models to evaluate hematoxylin and eosin-stained slides would help identify this subset.

In this study, the researchers used hematoxylin and eosin-stained tissue sections obtained from hysterectomies performed on patients with endometrial cancer of the p53abn or NSMP subtype.

A discovery cohort of 368 patients was used in this study, and the results were validated using two independent cohorts of 614 and 290 patients.

The researchers also performed whole-genome shallow sequencing of representative samples of both subtypes from the validation cohort, as well as a sample of p53abn-like NSMP, and this data was used to analyze copy number and gene expression profiles.

result

The study found that AI-based analysis of histopathological images successfully identified a subset of patients with significantly poorer survival rates and more aggressive forms of cancer within the NSMP subtype.

This subset of malignant tumors accounts for nearly 20% of NSMP tumors and 10% of all endometrial cancers.

The results suggested that clinicopathological features, immunohistochemistry, next-generation sequencing molecular markers, and gene expression profiles may still not be able to distinguish p53abn subtypes from these p53abn-like NSMP cases.

The deep learning model also identified tumors that contained tumor proteins. TP53 Mutations were detected despite normal immunostaining for p53, which would have been a false negative based on immunohistochemical classification.

AI-based tools were able to identify a more aggressive subset of NSMPs in p53abn -like cancers, even when pathological and molecular features could not predict poor survival.

Shallow whole-genome sequencing showed that this subset of NSMP cases had a high proportion of altered unstable genomes similar to subtype p53abn, but the level of instability was low.

The findings of this study also provided evidence of histopathological differences in this subset, even though no pathological or immunohistochemical differences were observed with the NSMP subtype.

Conclusion

Overall, the findings show that AI-based image classifiers can distinguish between subsets of endometrial cancer patients and detect those with significantly poorer survival rates.

The researchers believe that this AI-based tool can be easily integrated into the clinical diagnostic process to routinely scan histopathological images.

Moreover, with further refinement, this AI-based tool could potentially replace more time-consuming and costly molecular marker-based diagnostic methods.