SL-IM project

Slide Imaging is a research project supported by DAS srl and developed by both the Università Campus Bio-Medico (Campus University Medical Informatics & Computer Science Laboratory) and the DAS company. It aims at providing a Computer-Aided Diagnosis (CAD) system for autoimmune diseases through the analysis of Indirect ImmunoFluorescence (IIF) images. Undeniably IIF is considered a powerful, sensitive and comprehensive test for antinuclear auto-antibodies (ANA) analysis besides being one of the most effective and widely-used diagnostic screening assays for the timely detection of some pathologies whose incidence has been constantly growing in recent years.

IIF slides are examined with a fluorescence microscope and their diagnosis requires both the estimation of fluorescence intensity and staining pattern description. The former is scored semi-quantitatively with both positive and negative controls contained in each slide. The latter suggests the localization of reactive nuclear antigens and may help clinicians in differential diagnosis.

However, IIF method has some disadvantages. The major ones are: the low level of standardization, the inter-observer variability which limits the reproducibility of IIF readings, the lack of resources and adequately trained personnel and lastly, the photo-bleaching effect which significantly bleaches tissues in few seconds. Such drawbacks affect diagnosis repeatability thus limiting procedure reliability. Published intra-laboratory variability has been estimated between 7-10%.

Indeed human ability to detect and diagnose disease through image interpretation is limited due to non-systematic search patterns and the presence of noise. In addition the vast amount of image data that is generated makes the detection of potential disease a burdensome task and may cause oversight errors.

Another problem is that similar characteristics found in some abnormal and normal structures may cause interpretational errors.

Automation may offer a solution to the growing demand of diagnostic tests for systemic autoimmune diseases, as in other areas of medicine.

The ability to automatically determine the presence of autoantibodies in IIF would enable easier, faster and more reliable testing.

For this reason, developing an effective Computer-Aided Diagnosis (CAD) system is an obvious medical necessity to potentially support physicians decisions and overcome current method limitations. Indeed CAD methods, which have definitely been proven effective in other contexts

 

(i)                  allow a pre-selection of cases to be examined enabling physicians to focus their attention solely on relevant cases, facilitating carrying out mass screening campaigns

 

(ii)                serve as second diagnostic readers, thus increasing physicians capabilities and reducing errors

 

(iii)               aid physicians in their diagnosis

 

(iv)              work as training and educational tools of specialized medical personnel.

 

Besides providing image acquisition and traditional image post-processing tools, CAD main functions also include automatic image classification.

The analysis of the literature in the field of ANAs detection reveals that a comprehensive CAD system in IIF is not yet available.

Therefore this research project aims at developing a system that addresses the limitations reported above to potentially improve both data management and standardization level, with particular reference to image acquisition and classification.

As the first step of this project we have validated the use of digital images in IIF practice, both in manual and assisted diagnosis, with the following outcomes described here. First, three classes of fluorescence intensity (positive, negative and border zone) allow the identification of the presence/absence of auto-antibodies and distinguishing negative samples can also contribute to eliminating most routine samples. Second, key staining pattern classes (homogeneous, speckled, rim, nucleolar) potentially facilitate choosing the most appropriate second level auto-antibody serum tests.

In fact current computer diagnostic tools are capable of recognizing three classes of fluorescence intensity for HEp-2 images, prepared at the 1:80 titer, whereas in the near future all five staining patterns will be classified and recognized. Future features include an error-reject option based on being able to vary the operating points to render any classification system more flexible and personalised and therefore suited to a range of operating scenarios. The system has been developed based on images acquired using grey-scale cameras; image collection via color cameras is currently under investigation.

Such computer tools are based on image processing and pattern recognition techniques that we have specifically developed and tailored for this task. In particular, the fluorescence intensity classifier decomposes the classification polychotomy into a series of dichotomies according to the one‑per‑class method. At this level, the application of different aggregation rules enable the variation of system operating points.

 

We hope that these tools will improve the standardization level of IIF procedures and facilitate sharing information in this area of research to be subsequently employed as effective diagnostic and learning tools for training medical personnel.

 

For further information on the project, please visit Slide Imaging On Web