B-335 Analytical Validation of Desorption Electrospray Ionization Mass Spectrometry Imaging for Classification of Thyroid Fine Needle Aspiration Biopsies

Le résumé fourni par la source

Abstract Background Clinical methods for preoperative diagnosis of thyroid nodules using fine needle aspiration (FNA) biopsies, including cytology and genomic sequencing, can be slow, costly, or provide limited diagnostic performance, leading to unnecessary surgeries. As an alternative approach, we developed a method using desorption electrospray ionization mass spectrometry (DESI-MS) imaging and statistical modeling to classify thyroid FNAs, achieving accuracies above 83%. Furthermore, we performed a rigorous analytical validation study evaluating the performance of DESI-MS for classifying thyroid nodules using hundreds of FNAs. Key metrics including predictive accuracy, sample stability, precision, analytical specificity, and analytical sensitivity were methodically investigated. Methods Thyroid tissue sections (127 benign, 111 malignant) and prospectively collected clinical FNA biopsies (111 benign, 59 malignant) were analyzed using a Waters Xevo G2-XS QTof mass spectrometer fitted with a DESI-XS source. Molecular profiles from the thyroid tissue sections and 77 clinical FNAs were used to build classification models using logistic regression with lasso regularization. The model’s predictive performance was evaluated on the remaining 93 clinical FNAs. Analytical performance metrics were evaluated using 641 mock FNA samples prepared in the laboratory and analyzed using our standardized DESI-MS method. Results A classifier for differentiating benign and malignant thyroid samples was built and evaluated on an independent set of clinical FNAs, yielding an overall prediction accuracy of 83.9%, with a sensitivity of 91.4% and specificity of 79.3%. We next evaluated the analytical performance of the classifier. The impact of storage condition on classification was evaluated, with accurate classification achieved for mock FNAs analyzed after being stored up to one day at ambient temperature, one week in the fridge, and two months in the freezer. The method precision, including within run, between run, and between technologist precision, was evaluated using mock FNA smears (n=25/tissue) from 8 different tissues. Mass spectral profiles from replicate FNAs analyzed on different days by different technicians were highly similar, with standard deviations of the average classification probability less than 0.1. To assess analytical specificity of our method and evaluate its ability to distinguish benign and malignant cells in the presence of blood, we prepared mock FNA smears in the lab and mixed with increasing amounts of blood (0%, 20%, 40%, 60%, 80% blood). Accurate classification was achieved with up to 60% blood, with evidence to suggest that our method could be accurate with up to 80% blood. To evaluate the analytical sensitivity of our method and determine how many cell clusters are needed for sample classification, we used data acquired from the clinical FNAs and systematically predicted on an increasing number of pixels from each sample (1, 2, 3, etc.). The model’s performance dropped when using fewer than three pixels for sample classification, indicating that a cutoff of at least three pixels of data, which is the equivalent of 3 clusters of cells, is needed for reliable prediction. Conclusion Here, a DESI-MS method for thyroid FNA classification was developed and analytically validated. With the addition of DESI-MS, unnecessary diagnostic surgeries could be prevented by providing improved preoperative thyroid FNA classification.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.