The system makes the diagnosis based on images obtained from a confocal microscope and on artificial learning techniques. These techniques use cases previously diagnosed by specialists at the Hospital San Joan de Déu to generate a totally automatic diagnosis system. The system that is developed has a diagnostic reliability of above 95%, and is therefore a valuable tool for assessing objectively any new treatment that may be developed for these diseases.
Deficiencies in the structure of collagen VI are a common cause of neuromuscular diseases with manifestations ranging from Bethlem myopathy to severe Ullrich congenital muscular dystrophy. The symptoms of these diseases include proximal and axial muscular weakness, distal hyperlaxity, joint contractures and critical respiratory failure that requires mechanical ventilation and dramatically reduces life expectancy.
Structural defects in collagen VI are associated with mutations in the genes COL6A1, COL6A2 and COL6A3. However, despite current gene sequencing technologies, diagnosis remains difficult. This is generally the case in diseases caused by dominant mutations, where one of the major proteins is not completely absent, and when the effect of a genetic variant on the structure of the protein may not be evident. Therefore, before any genetic analysis, the standard technique for the diagnosis of dystrophies relating to collagen VI is the analysis of fibroblast culture images.
Specialists consider several aspects of the images, such as coherence in the organisation of collagen fibres, the distribution of the collagen network, and the arrangement of cells in this network to identify potential patients. However, this assessment is only qualitative, and regulatory agencies will not approve any treatment (such as gene editing using CRISPR technology) without an objective methodology to assess its effectiveness. Therefore, there is an urgent need for methodologies to quantitatively monitor the effect of any potential new treatment.
The proposed system responds to this need. It resolves the problem of the lack of data for learning that is typical in rare diseases, indicates potentially problematic areas in diagnostic images, and provides a general quantitative assessment of patients’ condition.
This development has been reported in the paper ‘A Convolutional Neural Network for the automatic diagnosis of collagen VI-related muscular dystrophies’, published recently in the scientific journal Applied Soft Computing (Vol. 85, December 2019). The study originates from a JAE introduction to research grant, funded by the CSIC and given to the then researcher at the IRI, Adrián Bazaga, the first author of the paper, who is currently associated with the University of Cambridge and STORM Therapeutics Ltd.