Creating AI tools that reflect these nuances [a multimodal approach] is essential for delivering personalized care.
Written by Shivani Choudhary
It seems that there is no place we can look to today that artificial intelligence (AI) and machine learning models haven’t pervaded. They have been used widely for their predictive prowess and ability to generate innovative solutions but have yet to be adopted in the medical field due to the higher stakes. However, recent developments in algorithmic analysis in the diagnosis process could improve patient outcomes.
Dr. Pallavi Tiwari’s work at the Integrated Diagnostics and Analytics (IDiA) Laboratory for Precision Medicine and Machine Learning for Medical Imaging initiative demonstrates how these technologies are beginning to transform the way we detect and understand different types of cancer. Her research centers on using AI to analyze multimodal data: the wealth of information gathered throughout a patient’s medical journey.
In traditional clinical workflows, data like imaging, pathology slides, and physician notes often exist in separate “silos,” considered as isolated pieces of information instead of a holistic picture. Each offers valuable insights, but none fully capture the complexity of diseases on their own. By integrating these aspects, AI models can generate a more comprehensive understanding of a disease, enabling clinicians to see a holistic picture that would be difficult to identify otherwise.
Illustration of three “silos” in medicine: pathology slides, imaging data, and physician notes.
This multimodal approach is especially crucial for cancers, where timely decisions can alter patient outcomes. Diseases like glioblastoma, pancreatic cancer, and breast cancer are notorious for their heterogeneity – tumors can vary significantly from one region to another and from one patient to the next. That complexity directly affects prognosis and treatment response and is often linked with poorer outcomes. Dr. Tiwari emphasizes that creating AI tools that reflect these nuances is essential for delivering personalized care.
Rather than building abstract models, Dr. Tiwari’s lab works closely with clinicians to ensure each project begins with a specific clinical question. This ensures the solutions developed are not “black boxes,” but intentional, interpretable tools that physicians can understand and trust.
One of the most promising areas of Dr. Tiwari’s research is opportunistic screening, which is the practice of testing for diseases in a patient already seeking care for other issues. Her team is developing AI systems that can analyze routine scans – MRIs and CTs that are already part of standard care – to flag potential tumors. This capability could serve as a “virtual biopsy,” using imaging alone to assess whether a lesion is likely to be malignant, and reducing the need for invasive procedures. This also makes care more accessible in places without access to specialized imaging, which are often necessary for correctly diagnosing cancers.
Beyond diagnosis, AI plays a growing role in prognosis and treatment planning. Dr. Tiwari’s models aim to answer questions such as: Will chemotherapy help this patient? Is the tumor aggressive? Should the patient pursue standard treatment or a more intensive approach? The goal is to give clinicians accurate, timely insights that help guide more effective care. In long-term management, AI may also help predict if and when a cancer is likely to recur, allowing physicians to tailor follow-up screenings and interventions more strategically.
In many ways, the ultimate mission is democratization. Advanced imaging scans can offer detailed insights but are costly and not universally available. Dr. Tiwari envisions a future where sophisticated AI models can extract comparable diagnostic value from routine, widely available scans, making high-quality care more accessible.
Across every step of the cancer journey, AI offers a powerful new set of tools. Dr. Tiwari stresses that these tools must be intentional, transparent, and deeply aligned with clinical needs. With this approach, AI has the potential not just to assist clinicians, but to fundamentally improve the way medicine understands and treats disease.
