For a cancer patient, it can be challenging to predict what lies ahead. Numerous factors are taken into account, including the patient’s health and family history, the grade and stage of the tumour, and characteristics of the cancer cells. However, it is ultimately up to medical experts who conduct fact-based analyses who determine the outlook.
According to Faisal Mahmood, PhD, an assistant professor in the Brigham and Women’s Hospital Division of Computational Pathology, this can result in “large-scale heterogeneity.” According to him, patients with the same type of cancer may have quite diverse prognoses, with some being more (or less) accurate than others.
He and his team created an artificial intelligence (AI) algorithm to make a more objective – and possibly more accurate – judgement because of this. The team’s findings have been published in Cancer Cell, and the purpose of the study was to see whether the AI concept was feasible.
More precision could lead to better treatment outcomes because prognosis is a crucial factor in treatment selection, according to Mahmood.
According to him, “[this technology] has the potential to produce more objective risk assessments and, as a result, more objective treatment decisions.”
Developing The AI
Using information from The Cancer Genome Atlas, a public database of profiles of various tumours, the researchers created the AI.
Based on genetics and histology (a description of the tumour and how rapidly the cancer cells are anticipated to proliferate), their algorithm forecasts the course of cancer (using DNA sequencing to evaluate a tumour at the molecular level). Mahmood points out that while genomics is increasingly used, histology has been the gold standard for diagnosis for more than 100 years.
Both are currently regularly utilised for cancer diagnostics at significant cancer institutions, he claims.
The 14 cancer kinds with the most data were selected by the researchers to test the system. The programme produced predictions that were more accurate when histology and genomics were combined than when either data source was used alone.
The researchers discovered that the AI used additional markers, such as the patient’s immune response to treatment, without being instructed to do so. According to Mahmood, this might imply that the AI can find new markers that we aren’t even aware of yet.
Mahmood is certain that this technology will be utilised for actual patients someday, perhaps within the next ten years, even though additional study is required, including extensive testing and clinical trials.
He predicts that in the future, large-scale AI models will be able to take in data from a variety of modalities, including radiology, pathology, genetics, medical records, and family history.
According to Mahmood, the more data the AI can take into account, the more accurate its evaluation would be.
Then, we are able to continuously and objectively evaluate patient risk.