As Fakrul Islam Tushar completed his PhD in November 2025, it is a natural moment to reflect on a journey shaped by both technical rigor and intellectual growth.
From early on, his work tackled one of the fundamental challenges in medical imaging AI: how to evaluate systems in a field where even experts may disagree. In areas such as lung nodule detection, different radiologists can interpret the same scan in different ways—meaning that the very data used to train and test models is often uncertain. Addressing this required not only technical solutions, but also a shift in perspective. Rather than focusing solely on building better algorithms, Tushar’s work emphasized the importance of understanding whether those algorithms can be trusted in real clinical settings.
This focus naturally connected to his research on virtual imaging trials and in-silico cohorts. By developing evaluation frameworks that do not rely exclusively on limited clinical datasets, his work contributes to bridging a well-known gap in the field: the difference between strong performance in controlled benchmarks and reliable behavior in real hospital workflows. In doing so, it helps move AI in medical imaging closer to meaningful clinical impact.
A key influence on this trajectory was the VICTRE study, which demonstrated the potential of virtual imaging trials as a rigorous alternative to traditional clinical evaluation. Encountering this work early on helped shape Tushar’s research direction, eventually contributing to developments such as virtual lung cancer screening approaches (VLST). It is a reminder of how a single well-executed study can inspire entirely new lines of inquiry.
Beyond the research itself, the PhD experience also brought important professional and personal growth. Over time, Tushar’s approach evolved from prioritizing model development to critically examining evaluation, deployment, and reproducibility. One guiding principle stayed with him throughout: a result that cannot be reproduced is not a result, but a coincidence. This mindset became central to how he structured his work, from large-scale computational pipelines to experimental design.
Equally meaningful was the environment at CVIT. The culture of questioning assumptions and engaging in rigorous discussion helped sharpen his thinking, while the shared experiences of deadlines, late-night debugging, and conference travel built lasting friendships. These are the moments that define a PhD just as much as the publications.
Among many memories, one stands out: presenting in a large conference auditorium on behalf of a labmate, only to be asked a question he could not answer. Thinking quickly, he pointed to his advisor in the audience and suggested the question be directed there instead—an unexpected moment that brought laughter to the room and became a story to carry forward.
As he moves on from CVIT, Tushar is starting a new chapter as an Assistant Research Professor at the University of Arizona, where he is establishing the Tushar Lab. Building on his PhD work, the lab will focus on data-centric AI for healthcare, with an emphasis on real-world deployment, evaluation standards, and foundation model benchmarking. The goal is clear: to develop methods and infrastructure that ensure AI systems are not only innovative, but also reliable and trustworthy in clinical practice.
We wish Tushar all the best in this exciting next step and look forward to seeing the impact of his work continue to grow.
