Assessing Tissue Fixation Time and Quality with Label-free Mid Infrared Spectroscopy and Machine Learning

Objectives: This work investigates whether changes in a biospecimen's molecular composition from formaldehyde fixation drive changes in the mid infrared (MID-IR) spectrum. Our ultimate goal was to develop an analytical metrology that could be used to accurately determine the fixation time of a tissue sample as a surrogate to overall tissue quality. Methods: Multiple unstained formalin-fixed paraffin-embedded tissue samples were scanned with an MID-IR microscope to identify a molecular fingerprint of formaldehyde fixation. The fixation specific patterns were then mined to develop a predictive model. A multiple tissue experiment using greater than 100 samples was designed to train the algorithm and validate the accuracy of predicting fixation status. Results: We present data that formaldehyde crosslinking results in alterations to multiple bands of the MID-IR spectra. The impact was most dramatic in the Amide I band, which is sensitive to the conformational state of proteins. The spectroscopic fixation signature was used to train a machine-learning model that could predict fixation time of unknown tissues with an average accuracy of 1.4 hours. Results were validated by histological stain quality for bcl-2, FOXP3, and ki-67. Further, two-dimensional imaging was used to visualize the spatial dependence of fixation, as demonstrated by multiple features in the tissue's vibrational spectra. Conclusions: This work demonstrates that it is possible to predict the fixation status of tissues for which the preanalytics are unknown. This novel capability could help standardize clinical tissue diagnostics and ensure every patient gets the absolutely best treatment based on the highest quality tissue sample.