This Saturday, October 5 at the Wagram room in Paris, it was glitter, fluff and party favors. The French radiology community celebrated 20 years of organized breast cancer screening, with celebration of the pioneers behind this program, testimonies from women and thanks to the minister (Philippe Douste-Blazy) who, at the time, had enabled its start. But behind the smiles and congratulations, this anniversary leaves a bitter taste.
If radiologists do not have to be ashamed of their results, with a 20% reduction in mortality linked to these tumors, the difficulties cannot be avoided. Too low participation of the women concerned (only 47% of those aged 50-74), lack of involvement of some radiologists, long delays in certain areas, whether to obtain an appointment or receive the report of the examination… Much remains to be done to improve this prevention system, and some already imagine that the solution will come from artificial intelligence (AI).
The promises are there. “The majority of studies show that a radiologist assisted by an AI has better performance than a radiologist alone,” recalls Guillaume Herpe, radiologist at Poitiers University Hospital and medical director of Incepto, a start-up specializing in the deployment of AI solutions for medical imaging. From there to imagining that algorithms replace one or other of the stages of screening, there is only one step… which promises to be a long one to take. Many obstacles still stand in the way of its deployment, particularly in France. At issue: our administrative particularities, surprising technical problems, the specificities of screening in our country and, of course, some financing problems.
A quick reminder of the organization of French screening: women have their mammograms in an imaging center, most often private, where a radiologist makes an initial reading of the images. In the event of an anomaly, he carries out additional examinations, or refers the patient for a diagnosis. In the event that no abnormality is seen on first reading, the images are sent to a “second reading” center where a radiologist with extensive experience in breast cancer examines them again. Around 5% of tumors not detected at the first stage are identified by this second level of analysis.
A stupid problem with… pipes
A first solution to streamline screening would be to replace the second reading with AI. Experienced radiologists who agree to go to dedicated centers for this tedious and poorly paid work are indeed becoming a rare commodity. “Four euros per reread mammogram, when this replaces or adds to the activity of their own practice, it is not very attractive, especially since the medico-legal risk in the event of an error is very real “, confirms Brigitte Séradour, president of the association of regional cancer screening coordination centers. While health insurance recommends delivery within two weeks, it happens, according to data collected by Incepto, that deadlines exceed six weeks. In Paris last year, they even reached a four-month record.
But to assist or replace the second reader with AI, it will first be necessary to solve a stupid problem with… pipes. In most French regions, the sending of images to second reading centers is still not electronic. “The National Cancer Institute has been organizing experiments for ten years and for the moment, no solution has been found on a national scale,” laments Professor Isabelle Thomassin-Naggara, head of department at the hospital. Tenon (AP-HP) and president of the Women’s Imaging Society (Sifem), the learned society which brings together experts in the field. As surprising as it may seem, the radiologists who perform mammograms, although they are digitized and read on screens in their offices, continue to print them on film, then have them taken to expert centers so that they can be reread. Absurd as well as apparently insoluble.
Another problem, and not the least, the major validation studies of artificial intelligence in the context of breast cancer screening have all been carried out abroad. “However, in most countries, the organization of screening programs is very different from the French system,” notes Professor Thomassin-Naggara. In France, the first reading is done in the office, but abroad, it is often carried out directly in expert centers. “The first reader is therefore immediately an expert reader, very specialized in breast imaging, whereas in our country the first readers are often more general radiologists,” explains Isabelle Thomassin-Naggara.
New studies essential
It is therefore impossible, according to this expert, to transpose foreign studies to France where AI replaces the second reader since abroad, there always remains an expert reader, which would no longer be the case in our country. “It will be essential to carry out studies in France before starting,” insists the radiologist. The risk otherwise: a potential deterioration in the quality of screening, which would be to the detriment of women. “But to carry out these tests, it will first be necessary to dematerialize the image transfer, and to equip the second reading centers,” continues Isabelle Thomassin-Naggara.
Perhaps we could use AI to sort mammograms in advance based on a degree of risk assessed by the algorithm. Those classified as high risk would benefit from the usual double reading, and those classified as low risk would then be read by a single reader. But in the French context, we come back to the previous problems. The alternative, bypassing the first reading and sending the images directly to the review center, does not seem more suited to our model, where women appreciate being able to discuss with the radiologist.
Last hypothesis, that AI facilitates the work of the first readers, for example as an “assistant” to the radiologist. In fact, this is already somewhat the case. “Our data shows that there are already more than a million patients per year for whom an AI helps with interpretation, or 20% of the screened population,” indicates Guillaume Herpe, of Incepto. Going further would require convincing a larger proportion of radiologists to adopt AI. In the field, not all doctors are necessarily enthusiastic. “It’s very practical for securing our readings in case of fatigue during the day, but honestly, it is really very rare that the algorithm detects lesions that I would not have seen. In addition, there are still far too many false positives: it reports anomalies which are not anomalies, and I waste time checking”, summarizes Dr Clarisse Lafont, private radiologist in the 11th arrondissement of Paris.
This breast specialist is currently benefiting from the tool free of charge thanks to a partnership with a start-up. She is also not convinced that the economic equation is there. “The prices seem very high – a manufacturer has already offered me a price of 10 euros per patient. The colleagues in sector 1 (without excess fees, Editor’s note) will never be able to assume such a cost given the prices at which they are charged. paid,” she adds.
A financing problem
That’s the whole question. Who will pay for the equipment of radiologists and second reading centers, but also for the dematerialization of image flows? “We are going to open discussions with the radiologists, and all these subjects will be on the table,” assures Marguerite Cazeneuve, the deputy director of Health Insurance. In the short term, Social Security mainly wants to encourage radiologists to offer more slots to women. A question of money there too. If there is indeed a strong disparity in the distribution of specialists across the territory, with real difficulties of access in certain places, these are reinforced by another very French specificity: the cohabitation of organized screening with individual screening.
In the first case, women are fully reimbursed, and the doctor cannot request additional fees. In the second, professionals are free to request price supplements. Some therefore tend to favor these appointments over those of organized screening, which further reduces the slots available for the offer 100% covered by Social Security, and reinforces inequalities in access to care. However, women have every interest in turning to organized screening, which offers reinforced quality guarantees (double reading, statistical monitoring of performance, etc.).
As part of future discussions with professionals, Health Insurance therefore wishes above all to find a way to encourage radiologists to enroll more in this offer. “As part of a memorandum of understanding, there may be financial incentives, with in return a form of moral commitment to offer a percentage of appointments to be determined as part of organized screening,” continues Marguerite Cazeneuve .
It will then also remain to convince women to come. Health Insurance, which has taken over the management of invitations since this year, has put itself in battle gear to achieve this. Individual letters, personalized reminders, information from treating doctors on whether or not their patients participate in screening and soon from radiologists themselves: numerous measures are being deployed to ensure that as many women as possible are informed of this possibility. It’s then up to them to decide.
.