In this decision, the European Patent Office did not grant a patent on determining cardiac output by the aid of an artificial neural network. The decision gives some very relevant clues as to how specific the neural network, and its adaptation to the particular use-case, has to be described in the application. Here are the practical takeaways of the decision T 0161/18 (Äquivalenter Aortendruck/ARC SEIBERSDORF) of 12.5.2020 of Technical Board of Appeal 3.5.05:
- The present invention which is based on machine learning in particular in connection with an artificial neural network is insufficiently disclosed, because the training of the artificial neural network according to the invention cannot be carried out due to a lack of disclosure.
- Because in the present case the claimed method differs from the prior art only by an artificial neural network, the training of which is not disclosed in detail, using the artificial neural network does not lead to a special technical effect which could be the basis for an inventive step.
* (inofficial translation of the decision)
This European patent application generally relates to a method for determining cardiac output. Cardiac output is a term used in cardiac physiology that describes the volume of blood pumped by the heart per unit of time. The patent application describes a variety of known ways to determine cardiac output, all of them being invasive and thus “expensive, impractical and reserved to intensive medicine”.
Furthermore, the application describes it as known to perform a pressure measurement non-invasively in a peripheral region as an indicator of the cardiac output. However, the arterial blood pressure measured at the periphery is distorted as compared to the aortic pressure, thus requiring the peripheral blood pressure curve to be mapped onto a corresponding central blood pressure curve, i.e. onto the equivalent aortic pressure. Such a transformation of the blood pressure curve is, according to the patent application, extremely complex.
The patent application therefore aims to provide a method for determining cardiac output which, based on the arterial blood pressure curve measured at the periphery, enables the precise determination of the cardiac output, wherein computational expenditures are to be kept within reasonable limits in order to enable its integration in a mobile appliance.
Here is how the invention is defined in claim 1:
Claim 1 (inofficial translation)Method for determining cardiac output from an arterial blood pressure curve measured at the periphery, in which the blood pressure curve measured at the periphery is arithmetically transformed to the equivalent aortic pressure and the cardiac output is calculated from the equivalent aortic pressure, characterized in that the transformation of the blood pressure curve measured at the periphery into the equivalent aortic pressure is performed by the aid of an artificial neural network the weighting values of which are determined by learning.
Is it patentable?
The first-instance examining division had refused the patent application because it had found the independent claims to be obvious in view of the prior art. But the board of appeal took a step back and raised an objection under Art. 83 EPC (insufficient disclosure), in addition to lack of inventive step (Art. 56 EPC).
Insofar, the EPC requires that the European patent application has to disclose the invention in a manner sufficiently clear and complete for it to be carried out by a person skilled in the art. The board added that to this end, the disclosure of the invention in the application has to enable the skilled person to reproduce the technical teaching underlying the claimed invention based on his/her common general knowledge.
Here is how the board assessed the invention against these principles (the citations in this article are translations of the German decision):
The present application uses an artificial neural network to transform the blood pressure curve measured at the periphery into the equivalent aortic pressure. With regards to the training of the neural network of the invention, the application merely discloses that the input data should cover a broad spectrum of patients of different age, gender, constitutional type, health condition and the like, to avoid a specialization of the network. But the application does not disclose which input data is suitable for training the artificial neural network of the invention, or at least a data record suitable for solving the underlying technical problem. Hence, the training of the artificial neural network cannot be reproduced by the person skilled in the art and the person skilled in the art therefore cannot carry out the invention. The present invention based on machine learning in particular in the context of an artificial neural network thus is insufficiently disclosed, because the training of the invention cannot be reproduced due to a lack of disclosure.
As a result, the board decided that the application does not meet the requirements of Art. 83 EPC.
Turning to inventive step, the application aims at a precise determination of the cardiac output while keeping the computational requirements within reasonable limits. The board, however, decided that the claimed solution does not involve an inventive step:
Claim 1 solves this problem with the aid of an artificial neural network, the weighting values of which are determined by learning. The appellant argued that the use of an artificial neural network has the technical effect that the cardiac output based on the arterial blood curve measured at the periphery can be determined reliably and precisely taking into account the narrow-band nature and resonance phenomena in the low frequency range of the transmission path between the aorta and the periphery, wherein the computation efforts are kept within reasonable boundaries, which allows an integration into a mobile and handy device. The board is not convinced that the artificial neural network of claim 1 takes into account the narrow-band nature and resonance phenomena in the low frequency range of the transmission path between the aorta and the periphery, because neither the claim nor the description contains details about the training of the artificial neural network. The board holds that the mere mentioning that weighting values are determined by learning does not go beyond the normal understanding of an artificial neural network by the person skilled in the art. In the present case, the claimed neural network is thus not adapted for a specific claimed application. According to the board, only a non-specific adaptation of the weighting values is performed, which is part of the nature of every artificial neural network. Therefore, the board is not convinced that the argued technical effect is achieved in the claimed method over the whole range. This effect can thus not be considered in the sense of an improvement over the prior art when assessing inventive step.
Since the subject-matter of claim 1 does not lead to an improvement over the prior art, the objective problem is to provide an alternative to the method disclosed in the closest prior art. The solution of this problem (using an artificial neural network, the weighting values of which are determined by training) does not involve an inventive step. The use of artificial neural networks not only follows a general technological trend, it was also known for the transformation of the blood pressure curve measured at the periphery into the equivalent aortic pressure. […]
Therefore, claim 1 was also found not to involve an inventive step, and the appeal was dismissed.
You can read the whole decision here: T 0161/18 (Äquivalenter Aortendruck/ARC SEIBERSDORF) of 12.5.2020