Ken Hanson

I met Bob Wagner in 1976 at an SPSE meeting in Toronto. Bob and I quickly struck up a close friendship that has lasted ever since. I was new to the world of imaging and eager to learn what it was all about. Since we both had a background in physics, I found it easy to talk to Bob. He had gotten into medical imaging a few years earlier as a researcher with the Bureau of Radiological Health at the FDA and already had a grasp of some of the fundamental issues facing the field. He helped me understand the basics of imaging, just as he has helped so many others.

We soon found ourselves regularly talking to each other on the phone. In those days, medical imaging was mostly about x-rays, that is, radiography and computed tomography (CT). As physicists, we knew all about photons and realized that their use implied a fundamental limitation to medical images because of their quantum nature. The statistical nature of photons clearly would lead to noise in the images. The first issue to address was to characterize the noise in images by estimating the Noise Power Spectrum (NPS). The most fundamental question to ask was how close to “quantum limited” medical images were. We were led to embrace the concepts of NEQ (Noise-Equivalent Quanta) and DQE (Detective Quantum Efficiency), promoted for characterizing photographic images by Chris Dainty and Rodney Shaw, who incidentally were the organizers of the 1976 SPSE conference where Bob I first met.

Bob wrote several papers in the late 70s about how NEQ was related to the signal-to-noise ratio for detecting objects in radiographs. Out of our collaboration came a paper that we wrote with John Sandrik in which we demonstrated that the NPS of film samples could be consistently measured by different labs.

Immediately following the SPSE meeting, Bob and I drove to Ottawa together at attend the International Congress of Medical Physics. There we heard David Chesler, from the University of Wisconsin, talk about the noise in CT reconstructions and how it was anticorrelated. That talk had a profound effect on Bob’s future studies, and mine as well. We both became interested in characterizing the consequences of correlated image noise on image interpretation. I ended up measuring the noise power spectra (NPS) in reconstructions from commercial CT scanners; Bob, David Brown, Art Burgess, and I coauthored a paper on the consequences that x-ray projection imaging had on detectability in reconstructions. Bob went on to consider other x-ray detection geometries.

In 1981 my family drove across the country so our kids could see the East coast. We spent a few days with Bob; he even accompanied us to see some the Washington DC sites. One morning, he surprised us by offering blueberries with cereal for breakfast. From then on, my kids called him “the blueberry man.”

In the 80s and 90s, I made regular pilgrimages to the Center for Devices and Radiological Health, to visit Bob and his colleagues. I would try to spend a week there, because it was such a vibrant and friendly group. Bob and I enjoyed discussing a wide variety of topics. On our next meeting, our discussions seemed to pick up exactly where they had previously left off.

The SPIE Medical Imaging meeting was first organized in 1984 by Rodger Schneider and Sam Dywer. In their wisdom, they changed the direction the meeting to focus on imaging technology. Roger was “Bob’s boss.” I am sure Bob was influential in determining the new bearing of the meeting, if only by supporting Roger's inclinations. His relaxed manner contributed to making the meeting itself a great success. He helped promote an informal atmosphere, while maintaining at truly high intellectual level, by asking speakers non-threatening questions with the purpose of either clarifying what they had said, probing the consequences of what they presented, or occasionally suggesting a possible flaw in their approach.

In 1987 I began collaborating on image evaluation with Bob and Kyle Myers, who had recently joined the CDRH. We decided to seriously address the topic of how to compare images produced by different image-processing algorithms. Our idea was to evaluate images in terms of how well a specific visual task can be performed using the images. The particular modality we chose was CT. We wanted to evaluate the efficacy of CT reconstruction algorithms when faced with incomplete or severely degraded projection data. Nonlinear reconstruction algorithms seemed to produce visually pleasing results, but the real question for us was whether they were actually better for performing visual tasks. We considered the ‘ideal machine observer’ and drew connections to the performance of human observers, eventually to even employ neural networks. Coauthors in the series of papers we wrote on this topic, and its extensions, include Harry Barrett, Jannick Rolland, Dave Brown, and Mary Pastel Anderson.

In all, Bob and I were coauthors on eight papers. I thoroughly enjoyed working with him because with his openness; he always made it a true collaboration.

I will remember Bob for his irrepressible curiosity about science, and generous nature. He was always willing to help people understand difficult concepts. He loved his work and discussing new ideas, but he also enjoyed people and had a great sense of humor.

I loved Bob Wagner like a brother and I will miss him.

Ken Hanson, Los Alamos