Artificial intelligence for the diagnosis of rare diseases related to collagen VI


Visualisation of the diagnosis of a fibroblast culture image. Each patch of the image is individually assessed, which helps to quickly identify areas with defects in collagen VI. The system also provides a general diagnosis to track patients.


Fibroblast cultures seen through a confocal microscope, in which collagen network (green) and fibroblast nuclei (blue) are visible. Left: a control sample. Centre: the sample of a patient with Bethlem myopathy. Right: the sample of a patient with Ullrich muscular dystrophy.

Researchers at the Institute of Robotics and Industrial Informatics—a joint centre of the CSIC and the UPC—and the Sant Joan de Déu Barcelona Children’s Hospital have developed a system for helping diagnose rare diseases related to deficiencies in the structure of collagen VI.

Jan 21, 2020

The researchers Adrian Bazaga and Josep Maria Porta from the IRI—a joint centre of the Spanish National Research Council (CSIC) and the Universitat Politècnica de Catalunya · BarcelonaTech (UPC)—have developed a system based on artificial intelligence for the diagnosis of rare diseases related to deficiencies in the structure of collagen VI. Specialists from the Sant Joan de Déu (SJD) Barcelona Children’s Hospital have worked with researchers from the Instituto de Salud Carlos III (ISCIII) in Madrid, the University of Cambridge (United Kingdom) and STORM Therapeutics, a British technology-based company.

The system makes a diagnosis from images obtained with a confocal microscope and it is based on artificial learning techniques. These techniques learn from cases previously diagnosed by specialists from the SJD Barcelona Children’s Hospital to generate a fully automatic diagnostic system. The developed system achieves a reliability in the diagnosis superior to 95%. It could become a valuable tool to objectively assess any new therapy to treat these diseases.

A common cause of neuromuscular diseases
Deficiencies in the structure of collagen VI are a common cause of neuromuscular diseases with serious manifestations ranging from Bethlem myopathy to severe Ullrich congenital muscular dystrophy. The symptoms of such diseases include proximal and axial muscle weakness, distal hyperlaxity, joint contractures and critical respiratory failure that requires assisted ventilation, which dramatically reduces life expectancy.

Structural defects of collagen VI are related to mutations of the COL6A1, COL6A2 and COL6A3 genes. However, despite current genetic sequencing technologies, diagnosis remains difficult. This generally happens in diseases caused by dominant mutations, in which there is no complete absence of a major protein, and when the effect of a genetic variant on the protein structure may not be evident. Therefore, before any genetic analysis, the standard technique for the diagnosis of dystrophies related to collagen VI is the analysis of fibroblast culture images.

Specialists take into account several aspects of the images, such as the coherence in the orientation of the collagen fibres, the distribution of the collagen network and the arrangement of the cells in the said network to identify potential patients. However, this evaluation is only qualitative, and regulatory agencies will not approve any treatment (such as genetic editing using CRISPR technology) without an objective methodology to assess its effectiveness. Therefore, there is an urgent need for precise methodologies to quantitatively monitor the effects of any possible new therapy.

The proposed system responds to this need. Such a system is advantageous since it solves the problem of the lack of data for typical learning in rare diseases, points out the possibly problematic areas in the consultation images and provides a general quantitative assessment of the condition of patients.

The research is explained in the article “A convolutional neural network for the automatic diagnosis of collagen VI-related muscular dystrophies”, recently published in the scientific journal Applied Soft Computing (Vol. 85, December 2019). This research was carried out on a JAE introduction to research grant, funded by the CSIC and awarded to the IRI researcher Adrián Bazaga, the first author of the reference article, linked to the University of Cambridge and STORM Therapeutics.