AI-predicted cancer prognosis and treatment response

Till date, AI helped doctors review images detect disease-associated anomalies, but there was no computerised model to incorporate multiple data types. Consequently, AI tools primarily catered to diagnostics (detecting the disease) rather than prognosis (predicting the likely clinical outcome and identifying most effective therapy) This limitation posed a formidable challenge given that the fusion of visual information (like X-ray, CT, and MRI scans) and textual information (like exam notes, and multi-disciplinary cross talk transcripts) is integral to cancer care.

Now, Stanford Medicine researchers have developed MUSK (multimodal transformer with unified mask modeling) – a model which is likely to transform AI’s role in guiding cancer care.

It is pertinent to note that MUSK is a foundation model pre-trained on vast amounts of data that can be customised with additional training to perform specific tasks. The model’s ‘learning’ data pool is thus expanded by several orders of magnitude, and it can serve as a off-the-shelf tool, which doctors can enhance to make clinical predictions as accurate as possible.

Following its training on 50 million medical images of standard pathology slides and more than 1 billion pathology-related texts, the model outperformed standard methods to predict the prognoses of patients with diverse types of cancer; for instance, identifying which lung or gastroesophageal cancer patients could benefit from immunotherapy, and which melanoma patients are most likely to experience cancer recurrence.

Read more: