Machine learning revolutionises oncology, predicting risks and outcomes in challenging cancers for personalised care. Explore its transformative impact on clinical decision-making for improved patient outcomes.

We live in a time referred to as the “digital revolution.” AI, especially, has contributed to the rapid expansion of accessible health data, changing the way we practice medicine.

Improving precision oncology

Although there have been numerous advances in the diagnosis, prognosis, and treatment of cancer, individualised and data-driven care is difficult. This challenge is rooted in the distinct molecular, genetic, and tumor-based characteristics of the disease. 

“AI has become pivotal as it can provide patients with forecasting and prediction and improved risk stratification according to specific criteria, such as in the cases of some breast, colon, ovarian, lung, and skin cancers,” per the author of a review published in Frontiers in Public Health.[1] 

“Further, it can detect hidden patterns from several sources such as molecular profiling, pathology, and medical imaging, and integration of -omics data to provide a more comprehensive understanding of cancer and improve precision oncology.”

The author also notes that, during a surgical procedure, AI has been shown to detect and diagnose cancers in real-time.

More recently, algorithms have utilised microRNAs (miRNA) as a biomarker for cancer classifications. These miRNAs are small, single-stranded, non-coding RNAs that play a pivotal role in cancer pathogenesis.

Cancer of unknown primary

Between 3% and 5% of cancers are classified as cancer of unknown primary (CUP). The dearth of targeted therapies for CUP results in poor outcomes. 

The deep-learning tool was also able to identify CUP subgroups with significantly higher polygenic germ-line risk for the predicted cancer types, and with significantly different survival outcomes. Patients with CUP treated with first palliative intent treatments that were aligned with their OncoNPC-predicted cancers had better clinical outcomes (HR = 0.348).

Pancreatic cancer

Pancreatic cancer is usually caught late, thus resulting in poor outcomes. Dutch and US researchers trained machine-learning models on the sequence of disease codes in clinical histories of patients with pancreatic cancer and tested the prediction of cancer occurrence within incremental time windows.[3]

“We present a framework for predicting the risk of a low-incidence but very aggressive cancer by applying deep learning to real-world longitudinal datasets of disease trajectories,” the investigators wrote. “This study was designed to make explicit use of the time sequence of disease events and to assess the ability to predict cancer risk for increasing intervals between the end of the disease trajectory used for risk prediction and cancer occurrence.”

“Our results indicate that using the time sequence in disease histories as input to the model, rather than just disease occurrence at any time, improves the ability of AI methods to predict pancreatic cancer occurrence, especially for the highest-risk group,” they added.

According to the investigators, their model enhanced the ability to monitor patients at higher risk, potentially benefiting their lifespan and quality of life via early detection of this aggressive cancer.

But its potential extends beyond oncology: “These technologies also have the potential to improve the diagnosis, prognosis, and quality of life of patients with various illnesses, not just cancer,” said the authors. 

Disclaimer: This story is contributed by Naveed Saleh and is a part of our Global Content Initiative, where we feature selected stories from our Global network which we believe would be most useful and informative to our doctor members.