Fireside Chat: Toshio Takiguchi

Session Focus

Brink opened by noting that Canon is not widely understood as a healthcare company. The conversation traced how it became one, then examined the company’s validation philosophy, its work extending imaging to underserved populations, and how AI changes both the value equation for equipment and the economics of running a radiology department.

How Canon Became a Medical Company

Takiguchi traced the line back to the founder. Canon’s first president, Dr. Takeshi Mitarai, was a physician, and the company’s first medical device was an indirect X-ray camera developed in 1940 to conduct health checks on its own employees. Mitarai’s reasoning was that the company could only run if its people were healthy.

The subsequent steps: ophthalmology in the 1970s, applying the company’s optical technologies, then in the 1990s a medical component business that accelerated the shift to filmless radiography through a digital X-ray system with a large flat panel detector, which he described as a world first.

He called that period Canon’s medical business 1.0. Version 2.0 began with the 2016 acquisition of Toshiba’s medical business, making Canon a full-line medical equipment manufacturer.

That acquisition brought more than a century of imaging expertise, including real-time CT interventional imaging, helical CT, and a wide-detector CT system that he said revolutionized clinical utility particularly in cardiac CT. Toshiba also released the first commercially available superconducting MRI and has led ultrasound development for more than 50 years.

Brink added the clinical consequence from a radiologist’s perspective. The wide-area detector allowed a whole organ such as the heart to be imaged in a single rotation, which was a genuine revolution, and simultaneously created the crushing workload now facing radiologists, since one acquisition can generate five or six hundred images. Takiguchi’s response: we must do something for that.

Made for Life and the Validation Step

Takiguchi described made for life as Canon’s management slogan, committing the company to working with clinical professionals across three stages.

First, understanding the real clinical needs a product should address. Second, translating those needs into engineering solutions. Third, and the step most industrial products skip, validating outcomes and attendant value with clinical collaborators even when that is difficult and time consuming.

He acknowledged the approach may be called too conservative, but the company treats completion of that validation as a precondition for full global commercialization.

Partnerships supporting that work include Mass General Brigham, Johns Hopkins, Charité in Berlin and Nijmegen in the Netherlands, alongside numerous Japanese institutions. More recently Canon built an upstream development organization in Cleveland, co-located with new commercial headquarters and a Comprehensive Imaging Resource Center, supporting a partnership with Cleveland Clinic.

Reaching Patients Without Access

Takiguchi asked the audience to imagine being in the most remote accessible location in a country, and someone arrives by motorcycle or boat carrying a backpack X-ray device. It images the chest without connection to the grid or a generator, running on solar panels included in the kit. An AI application then determines whether the images need to go to a radiologist at all.

That is Delft Imaging, a Canon group company, and its CAD4TB tuberculosis screening solution in Africa, which combines screening with real-time surveillance, hotspot mapping and data insights supporting national health programs.

In the US, Canon has partnered with West Virginia University on a mobile lung cancer screening program serving rural parts of the state with limited or no existing access.

Brink connected this to his own work on the Lancet Oncology Commission, which documented the opportunity to extend cancer imaging care and estimated it could save 10 million lives over a decade, at a cost an order of magnitude below the return on that investment.

Using AI to Hold the Value Equation

Asked how a manufacturer manages the cost, regulatory and public pressure around AI, Takiguchi answered that every innovation, incremental or transformational, must prove its worth by easing operational burden, driving clinical efficiency or improving outcomes, and must be affordable.

He gave a specific engineering example. Developing next-generation CT, Canon built a 0.25 millimeter slice detector configuration that pushed manufacturing limits and cost. Image quality and resolution were superb, particularly for visualizing coronary plaque, but at that price point the configuration was only feasible at 4 centimeters of coverage rather than the 16 centimeters Canon is known for.

The solution was to train a deep learning model on data from the thin-slice system and apply it to the standard 0.5 millimeter, 16 centimeter scanner. That delivered 0.25 millimeter slice resolution on commercially available equipment at a sustainable price.

Brink translated for the non-imaging audience: this is having your cake and eating it too, using AI alongside existing hardware to double resolution without pushing the price beyond what the value equation supports. Takiguchi’s own summary was that the company did not want to go to market with a very high price.

His second example was the combined angio-CT system, a hybrid that marries CT with an interventional fluoroscopy suite so patients requiring both do not have to be moved between rooms. Canon began developing it in 1992, and it took more than 30 years to achieve global recognition, improving outcomes in trauma, oncology and cardiovascular applications. Each step required patience, resilience and partners with an appropriately long horizon.

From Narrow AI to Foundation Models

Brink framed the adoption problem directly. The first AI wave, beginning around 2016 when GPUs enabled massive parallel processing for narrow convolutional neural networks, saw slow adoption and few commercial successes. His diagnosis was friction: each narrow algorithm required radiologists to review an additional data set.

He positioned foundation models and agentic AI as the opening of a second wave with different characteristics.

On how AI justifies its cost, Takiguchi pointed beyond accuracy and downstream savings to staff retention, citing reductions in CT overtime exceeding 40 percent. His broader argument was that competitive advantage is shifting from building AI to operating it well.

Key Takeaways

1. Canon’s medical history runs to 1940. The founder was a physician, and the first device was built to check employee health.

2. Validation before commercialization is the differentiating commitment. Canon treats outcome validation with clinical partners as a precondition, not an afterthought.

3. AI can protect the value equation. Training on an unaffordable configuration delivered its resolution on affordable hardware.

4. Some innovations require decade-scale patience. Angio-CT took more than 30 years from development to global recognition.

5. Access work is operational, not aspirational. Backpack X-ray with solar power and AI triage is deployed today in Africa and rural West Virginia.

6. Operating AI well is becoming the advantage. Building models is no longer the scarce capability.