The state of Artificial Intelligence in medical imaging • Part I
Looking into the future with blinkers on

Rinckside 2022; 33,2: 3-4.

t a TRTF Round-Table Colloquium intrinsic and essential aspects of arti­fi­cial intel­li­gence were dis­cussed by a number of in­vited pro­fessio­nals in the field. This column and another one to follow will present some of the essential points that were touched upon.

First and foremost, there is no clear and generally accepted definition of what is con­sider­ed AI. One funda­mental outcome of the colloquium is the statement that there is no arti­fi­cial intelligence and never will be. There are “limited-intelligence” expert systems for dedicated applications, for instance aimed at computer-assisted diagnoses (CAD), chess play­ing soft­ware or self-driving cars — better described as brainpower and knowledge replacement software. But the chess software cannot knit and a self driving car cannot write novels. They cannot learn to knit and they cannot learn to write. Artificial diagnostic systems can diagnose a fracture but they can­not put the arm in a cast. These softwares will be permanent apprentices, biased and sub­ject­ive, never neu­tral and ob­ject­ive, re­flect­ing the ideas, ways of thinking and the input of their creators.

They are not transparent but in most cases com­plete­ly opaque. If you are a referring phy­si­cian and you want to know how the ra­dio­lo­gist came to a dia­gno­sis, the human image reader can explain it to you. Getting such an explanation from a machine will be difficult; it is unable to scrutinize and challenge the vera­city of the data it digests. Deep learn­ing AI cannot explain how it draws a conclusion — in particular if its “learning” is augmented with surrogate data collected from the inter­net. The number of trained radio­logists is shrinking. If there is no trained radio­logist around you have to live with the machine outcome: you have to believe its validity.

Algorithms can also be written in a way that the outcome is determined in advance by built-in bias, and certain procedures are re­com­mend­ed or even performed without further human de­libe­ra­tion and appro­val. Con­si­der­ing the state of the world one cannot trust a machine-intelligent system that is a black box. More so, in­creas­ing­ly, dol­tish and blun­der­ing di­let­tan­tes have access to research facilities — single-minded nerds, data autists — and unqualified “soft scientists”.

Opposition is building up against empty promises of what AI will be able to deliver.

Radiologists taking care of patients every day have a rather negative view of these nerds. Some years ago they would still con­sider computer geeks as part of aca­de­mia, but now they are placed into the drawer of “tech­ni­cians”. What used to be com­pu­ter or in­for­ma­tion science has lost its scien­ti­fic stand­ing and is simply in­for­ma­tics now, IT — the nerds are com­pu­ter or net­work tech­ni­cians. They meddle in medi­cal or scien­ti­fic questions with­out having any know­ledge or com­pre­hen­sion of prac­ti­cal medi­cine.

The technocratic attitude to develop novel data collection strategies and image re­con­struct­ion tech­ni­ques does not relate to deal­ing with sick people. It is part of a wild goose chase like many quantitative applications in medical imaging. Medicine is about human beings. The ad­vo­ca­tes of AI in medi­cine and parti­cu­larly in dia­gnostic imag­ing no lon­ger con­sider people. They are under the mis­con­ception that one can re­con­struct a living per­son using data: Humans are reduced to data-deliver­ing objects to be ad­mini­stered and pro­cessed by health care desk jockeys.

The emphasis of arti­fi­cial intel­li­gence is on a col­lect­ive rather than in­di­vi­dual de­script­ion. It works with sta­tist­ics, with averages. It’s assembly line health care, not the medi­cine that has been the ethical base of being a medi­cal doctor until a while ago. The idea­listic goal of per­sona­lized medi­cine is being trampl­ed on by the same people who pro­pagat­ed it as our goal some years ago.

AI will have the position of a middleman between medical doctor and patient, giving little but making a pro­fit for the manu­fact­urer. It will de­fini­tely be a major new cost factor in medi­cine, not only in de­velop­ment but also in main­te­nance costs. And there is no proof whatso­ever if the value of AI out­weighs the value of a train­ed me­di­cal doctor. Except if the me­di­cal train­ing in the rich countries gets even worse than it’s now. Is it Work in Progress or Work in Regress?

Citation: Rinck PA. The state of Artificial Intelligence in medical imaging • Part I
Looking into the future with blinkers on.
Rinckside 2022; 33,2: 3-4.

An abridged digest version of this column was published as:
AI: Is there a risk of looking into the future with blinkers on?
Aunt Minnie Europe. Maverinck. 15 April 2022.

TurnPrevPage TurnNextPage

Rinckside • ISSN 2364-3889
is pub­lish­ed both in an elec­tro­nic and in a prin­ted ver­sion. It is listed by the Ger­man Na­tio­nal Lib­rary.


→ Print version (pdf).

The Author


Rinck is my last name, and a rink is an area of com­bat or con­test.

Rink­side means by the rink. In a double mean­ing “Rinck­side” means the page by Rinck. Some­times I could also imagine “Rinck­sighs”, “Rinck­sights” or “Rinck­sites” …
⇒ more


Bulletin Board