Do we need more radiologists?
The world in these times is over saturating the maximum capacity of doctors – especially when compared to the growth of the population.
Europe is already a serious example. Without looking at the Latin American environment.
So much so that despite having greater resources, they run out of options in the face of a complex demographic transformation that is increasing the pressure on health resources throughout the world.
As the population grows, laboratory tests and diagnostic imaging studies (radiology) increase almost exponentially, and the gap is increasingly separated: between the capacity of the demand for body images (CT and MRI) versus the stagnation of radiologists trained available to report these images.
This is resulting in a significant increase in the workload for radiologists and is a manifestation of a challenge that is humanly unprecedented in radiology.
Governments and health institutions will have to address this as a matter of urgency.
To mention some specific cases, the lowest rate of radiologists per 100 thousand inhabitants is found in England, where there are 4.7 doctors with this capacity. It is these almost 5 specialists can not with the avalanche of studies, although they are considered as FTE that is to say as full-time radiologists (in English, full-time equivalent (FTE) radiologists).
The bottleneck for England, if it wants to close that distance to the demand, lies in a very ambitious goal of almost doubling its number of specialists to almost 8 per 100,000 inhabitants by 2022 according to the Royal College of Radiologists (RCR) (we are talking about FTEs !!, for what is considered already, an impossible, in fact for short or long term to cover the demand calculated for that future time, suffers a delay of almost 1,821 FTE specialists in images per year for every one hundred thousand inhabitants.
Well, France does not escape either. It is the country with the highest number of TACs per capita in Europe and that puts it even in a worse situation, where its specialists have an average of 51 years! and more than a third of those who practice today will not be in 17 years.
Concentrations in major cities: A failure of teleradiology
In spite of the benefits of telemedicine to be able to do readings of distance studies, which solves the availability of doctors of these characteristics in rural areas, it begins to be overtaken by two important phenomena in body image: one is the delay of the readings where the radiologists who make these interpretations are FTE by image centers in the main cities and read the studies with delay and less opportunity for the patients. In some countries the demand for reading has already surpassed them and the interpretation time has been falling.
The demand for TAC, for example, is increasing globally and this alone exceeds the capacity for human reading even with teleradiology. While the taking and reading of conventional studies (RX) has been decreasing except in countries such as Germany which maintains almost twice the average than its neighbors (with more than one exam per capita per year) and is much higher still in the taking of RM.
The other second effect created by the lack of opportunity in the reading -more not in the taking- is the value of the interpretation. Many studies show that there is no significant reading value for example in radiological orthopedic studies, since the diagnostic value especially in conventional readings is given by the orthopedist finally, displacing the written analysis of the radiologists, which has generated interest in the non-medical press as a justification for unnecessary spending to adjust the finances of high-cost equipment (In German: http://www.fr.de/wissen/gesundheit/medizin/schaedliche-strahlen-fuers-roentgen-braucht-man -einen-guten-grund-a-855836). Some journalists already ask themselves in other countries if the radiological study would have less cost by making the study or its interpretation unnecessary.
For this reason some already aim in the future to filter only body studies or requiring a second look – not to oversaturate the radiologists with normal studies – one of the leaders is the Swedish government that has already begun to take measures and plans to introduce reference guidelines, so that unnecessary examinations are not carried out and those that are carried out are the most appropriate for each case.
The behavior of demand and supply in radiology has shown in Europe, that the centers do not provide more supply and many of them already have additional hours payments of the same faculty members (80%), others opt for the famous outsourcing with non-faculty members (58%) or in the last ones called Medical ad-hoc locums (known as “hollow top”) that reach 48%.
That gap between demand and supply of reading has increased the number of radiologists ad-hoc locums ,,,
Because if the demand surpasses us in images, it is not viable to increase the number of graduates per year?
It is feasible to say: No more radiologists?
In principle no more of the same, that is, it is a mistake to increase the number of radiologists as is the market, because it would also increase the centralization of services in urban areas.
One reason for this phenomenon is the low rates of teleradiology reading, which are not comparable with the activities and high hospital complexity (includes interaction with specialists, medical boards, committees, etc.) of the FTEs in highly complex centers.
Another reason for some is the increase of FTE’s side jobs in readings with low opportunity in teleradiology in rural areas, which prevents the improvement in the opportunity in distant places, and as is already happening in other countries, the demand will surpass by far the offer of teleradiology regardless of the price elasticity.
If things go on, new graduates will be subjected to a low teleradiology work environment and will become Locums in the main centers of the most technified cities.
The birth of the Deep learning-radiologists (DLR)
“It is a mistake to hire many people to do complicated work. Numbers will never compensate for talent: two people who do not know how to do something are no better than one, it will slow down the process and the task will be even more difficult. ” Elon Musk
It is better to always think that a human when he can not with his capacity, think about the technology to be able to solve it. This technology is not an abstract concept of the eighties (when they were leading humanoid films in Hollywood) but refers to a reality to unify concepts in artificial intelligence (AI), to enhance human capacity, as has happened in so many other professions where Technology accelerates and multiplies the human limitation.
However, you have to unify concepts. When we talk about AI, we refer to a great project to build non-human intelligence, with some subdivisions that never require more human intelligence (which is outside the context of this analysis). Then within AI we can conceptualize Machine Learning (ML) where machines make us more intelligent in a very broad sense, so the concept of Deep Learning is born, born as a particular ML way for a specific task that is trained with us .
It is preferable to think in the short term in medical education in new radiology graduates being DLRs (making reference to the radiologists who will be practicing for the year 2025 !!) so that they arrive prepared from the postgraduate, dividing their traditional assistance training with a digital counterpart that Take them to be DLR:
what will allow them to understand technology as an ally and not as an adverse digital force that will take away their work.
Where you can reinvest your reading time, by avoiding almost the interpretation of normal studies (screening and routine imaging), leaving in the background the studies where their interpretation is not relevant, except when the trader requests his opinion in a required second opinion (as in the case of conventional X-ray studies that are assessed by orthopedists, without the need for a radiologist unless doubts arise). Therefore, the DLRs will be able to focus their efforts on readings of more complex cases (peer-review) on a large scale (taking into account the worldwide trend to decrease the conventional studies of RX (CR) and increase the body image (TAC / RM ).
To be able to make daily use of their own algorithms with AI, which do not require as much configuration each time as a residence visit to Algebra II (logarithms) to move from the so-called Traditional Computer Vision (TCV) to the new Convolutional Neural Networks (CNX) , where the engineering configuration is minimal, not to say almost zero, but which requires a greater architecture modeling (Garbage In / Garbage Out) which current radiologists are unaware of.
Use smaller sample sizes to test their algorithms in practice. We could think that for a consistent AI, the more data represent better forecasts, but the truth is that no; that is, a point is reached where the volume affects the measurements. For example, in general, 5,000 positives per type are required in order to have an adequate AI response. Then if we think that each radiologist can model their algorithms in the future to improve their reading quality and their number of studies easily surpasses that volume in less than 5 years. Seeing things like that, the goal of Deep Learning for radiologists at its best point: it will be to optimize the processes that perform more than ten thousand times (many times the ideal cut-off point is 12,500 …).
So it sounds plausible, that an image specialist or an institution after repeating a process 10k times can never process it again without AI intervention in Deep Learning to achieve the desired “zero false positives” that all radiologists dream of.
The DLR radiologists will not have to read at lower rates since the main centers of the system may offer reading from any other local or world center, using the algorithms trained by each institution.
Radiological interpretation is a fertile field for AI, but Radiology Departments and health institutions or leaders will wonder how to make it happen …
Well that’s a response that has to be adapted to each environment but this is a personal guide:
Deep learning-radiology (DLR): (CH + CM + PD) * ROI
Knowledge of human behavior (CH) + market behavior (CM) + predictions of diseases based on technology (PD) aka AI effect (today you can almost make mathematical predictions 48 hours before relapse occurs) high-cost patients, which leads to better forecasts and lower costs) x the reinvestment of cost optimization (ROI) (by releasing, for example, in the case of Colombia, 80% of expenses consumed by patients representing only 10%) % of users (of the 48 million Colombians (with more than 95% affiliated), 5 million consume 80% of the expenditure).
“I came to the conclusion that we should aim to increase the scope and scale of human consciousness to better understand what questions to ask. In reality, the only thing that makes sense is to fight for collective lighting. ”