A spectacular time saving in the emergency room for the management of fractures and help in the diagnosis of prostate cancer. Artificial intelligence is now part of the daily life of certain departments at Rennes University Hospital, which itself contributes to the development of several algorithms.
In total, the Rennes university hospital center already uses around thirty algorithmsartificial intelligence (IA), “including six in clinical routine”. One of the very first services to benefit from this technological contribution is the emergency department, which last year saw some 66,000 patients, around a third of whom were for trauma.
“In 2019, we realized that those who came for isolated limb trauma, such as a sprained ankle or a broken wrist, stayed on average 4 hours 45 minutes in the department,” says Dr Ulysse Donval, specialist in traumatology. to emergencies. How can we reduce this relatively long transit time without degrading the quality of care, and especially with constant staffing levels? It was AI that solved the equation with the integration of Boneview software into the hospital imaging systems, explains Dr. Donval.
This software – “extremely efficient in detecting fractures but also the absence of fractures”, he emphasizes – interprets the x-rays in just a few minutes. Concretely, the x-rays of the limb are displayed on the screen, with a yellow frame on the lesion that the algorithm thinks it has detected if applicable, and the mention “yes” or “no” in the “fracture” section, “ dislocation” and “effusion”. In case of doubt, linked for example to a splint interfering with the image, the software also indicates this.
Reduction of page time
Armed with this first interpretation, the emergency physician can then quickly see the patient, without waiting for a second reading by an often overwhelmed senior doctor, and allow him to return home if he does not have a serious injury. “All patients who leave the emergency room have their X-rays read” in the hours that follow by a radiologist, insists the doctor. And in the very rare cases where an anomaly has escaped the AI, the patient is contacted again for treatment.
The use of the software allows early discharge of patients without reducing the quality of care, with an error rate which has not increased, ures Dr Donval. This solution made it possible to reduce the average page time by 21% for patients without fracture and by 27% for those with fracture, a reduction of between 1 hour and 1 hour 20 minutes.
Diagnosis of prostate cancer
The Rennes University Hospital is no longer just a user but also a designer of AI, with the development of an algorithm to aid in the diagnosis by MRI of prostate cancer, the most widespread in France among men (more than 50,000 every year). However, this imaging technique for the prostate is “difficult to interpret” for non-specialist doctors, and it sometimes requires taking hundreds or even thousands of images, notes Dr Luc Beuzit. Hence the interest for this radiologist in “training” an artificial intelligence to read it.
In collaboration with the French start-up Incepto, Rennes University Hospital has created a database of some 6,000 prostate MRIs, half of which have been carefully annotated by Dr. Beuzit and a dozen of his colleagues. They then fed it into the algorithm. Since November 2022, this software, called Paros, has been used internally by radiologists on a daily basis. It automatically traces the contours of the prostate, calculates its volume and displays in red any suspicious lesion, which only remains to be confirmed by a biopsy.
According to initial essments presented to journalists on Tuesday, Paros is almost as accurate in his diagnoses as a senior radiologist and much better than an intern interpreting images alone. The software, soon to be certified, is about to be marketed throughout the world, rejoices the CHU.