Monitoring immunohistochemical staining variations by artificial intelligence on standardized controls

Quality control of immunohistochemistry (IHC) slides is crucial to ascertain accurate patient management. Visual assessment is the most commonly used method to assess the quality of IHC slides from patient samples in daily pathology routines. Control tissues for IHC slides are typically obtained from prior cases containing normal tissues or specific antigen-expressing disease samples, especially tumors. As such samples eventually run out, and tumors may be heterogeneous, constant expression levels from one control sample to the next cannot be guaranteed. With the increasing availability of standardized cell lines, the diagnostic use of these cell lines as alternatives to traditional laboratory-derived controls can be explored. Furthermore, stain quality of this cell line material can probably be better monitored with readout methods such as image analysis and artificial intelligence (AI) than with visual readout methods, in which accuracy is influenced by the training and experience of the pathologists and technicians. In this study, we present the results of our investigation into AI-measured stain quality of standardized cell lines designed as controls for HER2 and PD-L1 IHC stainings. Using 5 IHC autostainers from the same manufacturer and type, we quantified cell membrane expression levels of these cell lines after staining using Qualitopix, an AI algorithm for measuring stain quality control. Over a 24-month period of weekly AI measurements, we observed multiple unexpected variations, particularly in low- and medium-expressing cell lines. To further investigate these fluctuations, we assessed both interstainer variations and intrarun variations, revealing differences between the stainers and the slide slots within the stainers. To finalize our study, we performed HER2 and PD-L1 stainings on calibrator slides to measure the limit of detection to detect variance per stainer and slot. Our findings prompted extra maintenance from the manufacturer in one of the highly fluctuating stainers, which reduced variation. In conclusion, AI appears to be an effective approach to monitor the IHC stain quality of standardized control cell lines for therapeutic protein targets HER2 and PD-L1, and to trace the underlying errors. This may be crucial for accurate patient management.