Evaluation of Image Analysis for Interpretation of Her-2 Dual In Situ Hybridization (DISH)
Jeffrey L Fine, David J Dabbs, Kim McManus, Rohit Bhargava. University of Pittsburgh, Pittsburgh, PA
Background: Personalized breast cancer therapy includes evaluation for trastuzumab sensitivity, by testing for Her-2 gene amplification. Her-2 Dual in situ hybridization (DISH) is a test that combines visualization of Her-2 gene copies (in the context of chromosome 17 copies) with brightfield microscopy. Scoring Her-2 DISH is labor intensive, fatiguing, and slower than immunostain scoring. This is a great opportunity for automation and we report our evaluation of a newly available whole slide image (WSI) based image analysis (IA) system.
Design: Known Her-2 positive (22 TMA cores) and Her-2 negative (18 TMA cores) tissues were used. Her-2 DISH was performed per vendor's directions then slides were scanned as WSI at 40x magnification (iScan HT, Ventana, Tucson, AZ). Her-2 signal (black) and chromosome 17 signal (red) were scored using IA (Virtuoso, Ventana). If IA failed to auto-select nuclei, nuclei were manually circled; auto-selection was closely supervised and corrected as needed. IA errors were noted. Cases were eventually skipped if manual WSI preview showed low or no visible signal.
Results: Amplified cases were correctly identified in 100% of successful analyses (n=8), but only 47% of attempts were successful (n=17). Non-amplified cases were correctly identified in 12.5% of successful analyses (n=8), with 80% of attempts yielding a result (n=10). Many errors appeared to represent inability to "see" signal on IA, especially red signal. IA found nuclei automatically in 4 analyses and manual selection was performed in 23 analyses (n=27). In almost all cases, fewer than 10% of all selected nuclei generated a count result. 5 ampified and 8 non-amplified cases were not attempted after preview of WSI (inadequate signal).
Conclusions: True automation in surgical pathology remains elusive. IA had difficulty "seeing" low levels of signal, especially red signal. This caused over-calls, errors, and failure to auto-select nuclei. WSI did not compare favorably to 60x microscope image quality in terms of resolution and dynamic range. These issues were exacerbated by use of TMA cores, which contain fewer tumor cells and do not allow for individual pre-treatment of tissues. Additional validation using tissue sections may have better results. Numerous opportunities for workflow improvement were noted, especially when correcting or supervising nuclei selection. Despite these shortcomings, WSI-based interpretation of Her-2 DISH is highly desired; automation aside, the onscreen format is less fatiguing than microscope-based review and also potentially permits side-by-side comparison with the H&E slide.
Wednesday, March 6, 2013 1:00 PM
Poster Session VI # 275, Wednesday Afternoon