---

   Go to the List of Contents

The Latest Column

The State of Affairs
Impressions of 2025

Rinckside 2025; 36,1.

s times change, less and less people read columns like Rinckside be­cause the pre­vious­ly existing audience has changed — in most cases the attention span of today's readers has shortened dras­ti­cal­ly. Columns such as Rinckside are too long, too thorough, too de­mand­ing. They require cri­ti­cal think­ing as well as the ability and the men­ta­li­ty to fol­low argu­ments. Nowa­days, many readers are not used to this any more. They think in ste­reo­types and want short and simple "fast mental food" explanations.

That can be delivered — and is being delivered for instance by the "social me­dia"; tidbits of in­di­vi­dual statements mostly affected by personal, po­li­ti­cal, religious, commercial, in any case unobjective thinking to brainwash those people who fall for it. Many people accept such "news". The same holds for teaching websites and blogs. Some of the purveyors pro­bab­ly believe that they distribute true and im­por­tant information; however, most want to sell their opinions or peddle their ware.

Teaching Websites and Blogs

There are positive and well meant — and well done — examples, for in­stan­ce some chap­ters of the "European Society of Ra­dio­logy Modern Ra­dio­lo­gy eBook"; pros and cons of such contributions were dis­cuss­ed in a re­cent Rinck­side column:

⇒  Tablets versus textbooks — Back to the established roots.

The conclusion of this column was: Let's go for printed textbooks, in our case Magnetic Resonance in Medicine. But then reality struck — the question was: do these books reach the au­dien­ce we want to train? The answers was: no. Distribution is difficult and the price of such a book is too high, in par­ti­cu­lar out­side the European Union. Free-access websites seem to be the only way to reach the audience that needs this kind of education material, as easi­ly vi­sib­le in the readership maps.

Interestingly, after the last up­grade and up­date we counted more than 500,000 pageviews, within the last years even 1.5M views. We guess the rea­son is that people know that they can trust the content — it's neither ar­ti­fi­cial intelligence-based nor does it have any commercial background. We do not conceal information in order to achieve specific goals, and we se­pa­ra­te scien­ti­fic facts and opiniated comments.

Lower Requirements at Medical Schools.

Meanwhile another sign of decay has reached the medical schools: At the Ka­ro­lins­ka In­sti­tute in Solna, Sweden, the medical pro­gram lowers the re­quire­ments. Ellinor Kenne, coordinator of basic science courses taught to me­di­cal students at the start of their studies, remarked on the Vi Lärare web­site on 19 November 2024:

"We've changed a text­book for the third semester because it contains less text. We are lowering our ambi­tions a little in terms of the amount of text students have to read. There is no point in aiming too high, because then you lose stu­dents … Some people find it dif­fi­cult to read long texts, they would rather watch recorded lectures in­stead of read­ing course books."

The Swedes are not the only ones. A German university professor com­plain­ed in a news­pa­per inter­view that even reading moderately difficult texts is dif­fi­cult for the majority of the students.

Education in medical imaging and on-the-job training are com­pre­hen­sive pro­cesses that en­com­pass not only the acqui­si­tion of know­ledge and skills, but also the human and pro­fes­sio­nal de­ve­lop­ment of radio­logists. It fos­ters critical thinking, creativity and the abi­li­ty to navigate the com­plex world of medical imaging.

It also includes experiences, interactions and the pursuit of knowledge in every­day life and plays a crucial role in medicine by empowering people to make informed decisions and to ac­ti­vely parti­ci­pate in interdisciplinary me­di­cal diagnostics and treatment of patients. Judgment and the cor­re­spond­ing abi­li­ty to deal with risks and un­cer­tain­ties are pre­re­qui­si­tes.

Artificial Intelligence Lies in Wait Everywhere.

The whole field of medical imaging is changing — and undergoing further change for in­stan­ce by Applied Ar­ti­fi­cial In­tel­li­gence (applied AI). We are told that its algorithms and collected data can sup­ple­ment or even replace cus­to­ma­ry image reading by a trained ra­dio­logist. Nowa­days, all the cutting edge equipment is incorporating advanced in­for­ma­tion pro­cess­ing. It is ex­pen­si­ve and increases costs in the healthcare sector, and no­thing is stable, no­thing is really reliable. But apparently it is "politically and commercially cor­rect".

Reality is sobering, for instance:

"GPT-4V, in its current form, cannot reliably interpret radiologic images. Its ten­den­cy to disregard the image, fabricate findings, and misidentify de­tails, especially without clinical context, may misguide healthcare pro­vi­ders and put patients at risk." [Huppertz et al. European Radiology (2025) 35: 1111–1121].

or:

"Performance of AI was inferior to human readers in our unit. Having missed a significant number of cancers makes it unreliable and not safe to be used in clinical practice. AI is not currently of sufficient accuracy to be con­si­der­ed in the NHS Breast Screen­ing Pro­gramme." [Puri et al. Clinical Ra­dio­lo­gy. 15 March 2025. doi: 10.1016/ j.crad.2025. 106872].

A professor at the Technische Universität München made the following ab­surd state­ment to Germany's Alexander von Humboldt Foundation: "The huge ad­van­tage of AI models in medical imaging is that they always give you an answer, no matter how many images you put in front of them."

But we don't want "an" or any answer; we want the (correct) answer. Doesn't he realize that often the data quality of the input is bad and the necessary trust­wor­thy in­fra­struct­ure does not exist or requires a much greater tech­nical effort than ex­pect­ed? In many in­stan­ces the complexity of the problem to be solved is not taken into account by the promoters of the application nor by its users because they don’t understand the first thing about it.

Validation is a neglected or simply ignored factor. In the 1980s and 1990s I led an image pro­cess­ing group in the department I headed; a number of im­por­tant innovations in the field of image processing, image visualization, data collection, and early applications of very specific AI were developed dur­ing this time and became basic and expert know­ledge, including the knowledge of pitfalls and setbacks.

Validation is among them; it seems nearly impossible, because the pa­ra­me­ters of most di­gi­tal radio­lo­gi­cal exa­mi­na­tions are not exactly re­pro­duc­ible. However, extremely tho­rough va­li­da­tion must take place before AI algorithms are clinically feasible.

⇒  The question is of how we validate data.
⇒  It is frightening how data are collected and trained, even in powerful AI systems.

People don't know the basics. How certain can we be that training data may not contain an a priori error?

Still, it is a cash cow for developers and the industry. When I mentioned this some years ago, the article was not published because companies involved in ap­plied AI might with­hold their ad­vertise­ments. Today we know better — it's ra­ther expensive and traps medical imaging in a cycle of digital dependence.

In a follow-up of an earlier article, the British newsmagazine The Economist re­por­ted recently:

Last December, Aidan Toner-Rodgers, a 26-year-old PhD student at MIT, pub­lished a sensational paper titled, «Artificial Intelligence, Scientific Dis­co­very, and Pro­duct In­no­va­tion». Toner-Rodgers made extraordinary claims that at a large US firm, AI-assisted re­search­ers «dis­co­ver 44% more ma­te­ri­als, resulting in a 39% in­crease in patent filings,» and that «AI au­to­ma­tes 57% of 'idea-generation' tasks.» Daron Acemoglu, last year's No­bel Prize win­ner in economics, praised the paper.

In a footnote, Toner-Rodgers thanked Acemoglu for his «support and guidance.» The paper caused quite a stir and was covered by several publications including Na­tu­re, Wall Street Journal, and The Economist. It was cited by the Eu­ro­pean Cen­tral Bank and in the United States Congress. The problem is that Toner-Rod­gers had fabricated all the data. The paper was a fraud and had to be with­drawn.

Information Processing in Diagnostics and Therapy.

Many radiologists these days seem to have limited under­stand­ing of the equipment used for imaging and why, as well as how, it does what it does. Often they cannot read plain (and in­ex­pen­sive) x-rays any more. Everything is automated and in many cases radio­graphers take care of all scanning mat­ters. How long before we have robots doing their work? The re­ferr­ing phy­si­cians don't care, because they also lack the medi­cal and dia­gnostic back­ground.

Radiology as an independent medical discipline seems to be in a gradual de­cline. The de­mo­tion of radio­lo­gists to the rank of con­sult­ing imag­ing tech­no­lo­gists is a real threat to ra­dio­logy and the sooth­ing and appeas­ing remarks of some "opinion leaders" have not brought any moves to stabilize the si­tu­a­tion.

⇒  The Radiologist — rise to fame and slow fall.
⇒  Some thoughts about AI in medical imaging.
⇒  Some more thoughts — and basics — about AI in medical imaging.

Scientific papers in medical imaging seem of no consequence most of the time — it's not only journals; the European Congress of Radiology 2025, for instance, happily an­nounc­ed 10,000 abstracts; a (very) great number of them are irrelevant to clinical applications — or even non-medical. For the meet­ing in spring 2026, they invite: "Submit your abstract and help us beat last year’s record-breaking submissions."

Reading "scientific" journals gives me the creeps. Many papers seem to be written just to have a publication — with up to twenty or even more un­author­itative co-authors. They are in­com­pre­hen­sible and have no ra­dio­log­ical or medical relevance, playing around with AI soft­ware pro­grams. Re­cent­ly I was asked if I could proof­read a “good” paper: it was a meta-analysis of meta-analyses: what's the point?

The general erosion of the quality and of the background of scientific topics after the year 2000 is clearly perceptible in thousands of "scientific" papers.

⇒  What one expects from scientific publications.
⇒  Some thoughts about mass meetings such as ECR.

There is a quotation, several times repeated in Rinckside columns over the years:

“It is too much knowledge which leads to ignorance, because from a cer­tain moment on people only see the calculable part of things. And the har­mo­ny of numbers becomes their god … Progress makes the world in­creas­ing­ly smaller for people.

“And one day when people will be able to fly one hund­red miles a minute, the world will appear to them microscopically small, and they will feel like a spar­row on the top of the highest mast of a ship, and they will bend over to in­fi­ni­ty … and they will hate the machines which have turned the world into a hand­ful of digits and destroy them with their own hands.”
 
Guareschi G. Mondo piccolo, Don Camillo. Milan: Rizzoli 1948. The little world of Don Camillo. New York: Pellegrini and Cudahy, 1950.

Will we be faced with a kind of Frankenstein AI and see what Frankenstein's creator saw:

“I had desired it with an enthusiasm that far exceeded moderation; but now that I had finished, the beauty of the dream vanished, and breathless horror and disgust filled my heart … I looked at the monster — the miserable monster whom I had created."
 
Mary Wollstonecraft Shelley. Frankenstein; or, the Modern Prometheus. Chapter 5. 1818.

---

After 35 years of Rinckside, there won't be any new regular columns on this web­site, but you will be able to read occasional thoughts.

And, many of the old columns are still "current affairs" and will be in the future … and hardly anybody dares to mention the topics discussed in them. Check them out and write to me if you want to comment.

---