A global AI model for the diagnosis by image of renal anomalies developed in Japan and validated in Spain | NTT DATA


A global AI model for the diagnosis by image of renal anomalies developed in Japan and validated in Spain...

A global AI model for the diagnosis by image of renal anomalies developed in Japan and validated in Spain will boost clinical decisions support in hospital settings.

NTT DATA has announced the finalization of the AI assessment project for the classification of kidneys and detection of kidney anomalies which provides a useful support tool as a Clinical Decision Support System (CDSS) in hospital settings.

This project consists of the retraining and evaluation of AI models using abdominal CT scans provided by the Fundació Puigvert, a hospital which has provided around 3000 imaging studies, and has participated in the validation process of the observations made by the team of US radiologists. It forms part of a wider NTT DATA project based in a sample of 10,000 studies in the US.

The renal classification models are capable of classifying and distinguishing normal kidneys from kidneys with anomalies, for example, kidney masses, cysts, hypoplasia, hydronephrosis, scars and kidney stones.

The aim of involving several geographic areas in the AI training process responds to the need for the model to be built in representation of the different screening protocols used in different regions and to allow the inclusion of patients from different backgrounds and races in order to achieve universal validity of its applicability. In this collaboration between NTT DATA and Fundació Puigvert, AI systems have been retrained with thousands of images of different anomalies that can affect the kidneys, achieving high performance metric. Specifically, for the classification model, the F1 score, which summarizes the predictive performance of AI by combining two essential metrics - accuracy and recall - is 92.7% for the left kidney and 90.4% for the right kidney. These values represent an improvement of the performance of AI in comparison to results gathered from the original models on data provided by the Fundació Puigvert - 91.1% for the left kidney and 90.1% for the right kidney.

José Ignacio Aznar, leader of this project and head of Health Innovation projects at NTT DATA, explained that “it will be highly precise tool to improve and enhance efficiency in the diagnosis of renal anomalies, facilitating - in the long term - faster, safer and more reliable diagnosis".

Once this model is established in the workflow of healthcare professionals, it will help to lighten the workload and facilitate the detection of anomalies based on a global evaluation of the kidneys, in areas where it will have a great impact such as the emergency room, where in the medium-term, the CDSS with an AI core will be able to operate 24 hours a day.

The involvement of Fundació Puigvert

As a leader in the field of urology and nephrology in our country, the Fundació Puigvert has been involved since the beginning, providing the necessary resources for the identification and selection of abdominal CT scans to conduct the QA -Quality Assurance- of the observations. It has provided around 3000 CT scans, allowing the performance of the previous AI model to be improved with data from a region different to the original region. The center has also been actively involved in the definition and deployment of the infrastructure for the proof of concept and in the development of the deliverables associated with the results.