Breakthroughs in AI from the Bagci Lab Improve Diagnosis and Treatment
The Northwestern Radiology Department has been instrumental in developing ways to use artificial intelligence in imaging. Recent conversations with people in the Bagci Lab have yielded impressive updates in their work in this area.
Dr. Hatice Savas and Dr. Ulas Bagci’s collaboration at Northwestern has revolutionized their approach to medical imaging. Initially working on lung fibrosis scoring through visual assessment, their partnership shifted towards utilizing AI resources for more efficient processing of images. Dr. Savas highlights the transition from assessing hundreds to thousands of images, thanks to Dr. Bagci's expertise in AI.
Their research extends to an NIH-funded project titled "Predicting Post-Covid Pulmonary Fibrosis with Explainable Deep Learning and Optimal Biomarker Discovery." Additionally, they delve into oncology imaging, particularly focusing on prostate molecular imaging. Together with Dr. Yuri Velichko, they are engaged in an industry-sponsored study optimizing treatment and multi-imaging response evaluation for high-risk prostate cancer, employing PET, AI, and Radiomic data.
Dr. Savas emphasizes their commitment to addressing clinical inquiries through AI solutions. By collaborating closely with urologists, they leverage AI to tackle complex questions that traditional imaging methods may struggle to answer. Their goal remains steadfast: identifying clinical challenges, partnering with clinicians, and leveraging AI to enhance patient care.
Dr. Bagci reports that a 2024 breakthrough was the development of foundational models which is a new algorithm. Foundational models use large-scale data, including images of all kinds: texture, clinical, etc.
Dr. Bagci said, “Before, most of the algorithms were single modality. Clinical variables got you clinical analysis, and doctors had to choose between using image-based AI, clinical–based AI or pathology-based AI. With foundational models, the algorithm considers everything. AI analyzes all types of variables the way a doctor does. It solves diagnostic and prognostic problems, so we are able to predict patient outcomes much better than before.”
In 2019 the Bagci Lab developed the first algorithm for identifying pancreatic cysts which have a high likelihood of turning into pancreatic cancer. Its accuracy rate was 60% compared to humans’ 40-50% accuracy rate. As of 2024 with newer AI technologies, the accuracy rate is above 80%.
A priority of the Bagci Lab is increasing the intelligence level of current systems while keeping doctors at the center of the process of analysis. Dr. Bagci stresses the importance of radiologist-centered or human-centered AI. He says, “We have the technology to have AI learn from humans and to have humans learn from AI, in true collaboration.”
Frank Miller, Raj Keswani, and Gorkem Durak's work in the Bagci Lab has focused on using AI for pancreas segmentation used in the quantification of various pancreatic diseases. A recent breakthrough of the team was achieving automatic pancreatic segmentation from MRI and tomography. This function became publicly available in 2024, allowing anyone to download from it and to add their information to the data.
Also in 2024, Elif Keles, MD, PhD clinical fellow in the Bagci Lab, advanced the use of artificial intelligence for pediatric brain tumor segmentation and quantification, providing crucial insights into tumor size, volume, edema, and fluid accumulation. By accurately identifying four distinct tumor subregions, the AI model will facilitate individualized therapy selection, helping clinicians determine the most effective approach—chemotherapy, immunotherapy, or surgical intervention—tailored to each pediatric patient.
Dr. Keles said, "Our AI-driven model enables precise volumetric analysis and subregion classification, allowing us to optimize treatment strategies and improve clinical outcomes."
This pioneering work integrates AI-driven quantification with pediatric neuro-oncology, enhancing personalized treatment planning and refining clinical decision-making for pediatric patients.
Dr. Bagci’s team is now working on an algorithm to detect patient outcomes related to liver cirrhosis. They have segmented 600 MRI with different states of cirrhosis. Dr. Bagci works with world-class researchers from Northwestern, including Dr. Amir Borhani and Dr. Danila Ladner.
AI is becoming a bigger part of medical research and clinical work, and at many universities AI is a separate department, such as Stanford’s Institute for Human-Centered Artificial Intelligence. Northwestern’s AI researchers include clinicians who collaborate with The Bagci Lab to improve patient care with the latest breakthroughs.