The surgical intelligent knife distinguishes normal, borderline and malignant gynaecological tissues using rapid evaporative ionisation mass spectrometry (REIMS)

Abstract

Background: Survival from ovarian cancer (OC) is improved with surgery, but surgery can be complex and tumour identification, especially for borderline ovarian tumours (BOT), is challenging. The Rapid Evaporative Ionisation Mass Spectrometric (REIMS) technique reports tissue histology in real-time by analysing aerosolised tissue during electrosurgical dissection.

Methods: Aerosol produced during diathermy of tissues was sampled with the REIMS interface. Histological diagnosis and mass spectra featuring complex lipid species populated a reference database on which principal component, linear discriminant and leave-one-patient-out cross-validation analyses were performed.

Results: A total of 198 patients provided 335 tissue samples, yielding 3384 spectra. Cross-validated OC classification vs separate normal tissues was high (97·4% sensitivity, 100% specificity). BOT were readily distinguishable from OC (sensitivity 90.5%, specificity 89.7%). Validation with fresh tissue lead to excellent OC detection (100% accuracy). Histological agreement between iKnife and histopathologist was very good (kappa 0.84, P < 0.001, z = 3.3). Five predominantly phosphatidic acid (PA(36:2)) and phosphatidyl-ethanolamine (PE(34:2)) lipid species were identified as being significantly more abundant in OC compared to normal tissue or BOT (P < 0.001, q < 0.001).

Conclusions: The REIMS iKnife distinguishes gynaecological tissues by analysing mass-spectrometry-derived lipidomes from tissue diathermy aerosols. Rapid intra-operative gynaecological tissue diagnosis may improve surgical care when histology is unknown, leading to personalised operations tailored to the individual.

Conflict of interest statement

Z.T. serves as a paid consultant for Waters Corporation. J.B. is employed by Waters Corporation. The remaining authors declare no competing interests.

Figures

Fig. 1

REIMS set-up ex-vivo and in operating theatres (in-vivo) and frozen sample work-flow. a Electrical current, produced from the generator, is applied to the tissue and the resultant charged particles are extracted through the custom-designed hand-piece and drawn into the REIMS atmospheric inlet and analysed in the Xevo G2-XS mass spectrometer to produce tissue-specific mass spectra, which are then subjected to multivariate statistical analysis using PC-LDA. Within one to two seconds, real-time tissue diagnosis is displayed on a screen for the surgeon to see. b Work-flow for the frozen samples; all samples collected from the frozen tissue bank were processed with the iKnife. After histopathology reporting 22 samples were rejected from the study due to not being gynaecological or epithelial ovarian samples, or the histology was unclear and they failed quality control (QC). The remainder of the samples (n = 150) and resultant spectra were included in subsequent models, univariate and multivariate analyses

Fig. 2

Multivariate analyses of ovarian cancer and normal tissue types. a 3-component 3D PCA, percentage of variance explained in PCs1–3: 36.0, 17.6, 9.0%. b 25-component 3D PC-LDA. c Box plots of univariate analysis showing the top m/z peaks contributing to class separation (log2 fold change), between ovarian cancer and normal samples combined. Ten most significant and intense REIMS spectra peaks shown with P < 0.001, q < 0.001. Figure based on the following sample numbers: ovarian cancer n = 39, normal ovary n = 15, normal fallopian tube n = 15, normal peritoneum n = 14

Fig. 3

Multivariate statistical analyses of ovarian cancer, benign and borderline ovarian tumours. a 2-component PCA, b 25-component 2D PC-LDA, c Leave-one-patient-out cross-validation when all three tissue classes are included in the model, d Box plots of univariate analysis showing the top m/z peaks contributing to class separation (log2 fold change), between ovarian cancer and borderline ovarian tumours. Ten most significant and intense REIMS spectra peaks shown with P < 0.001, q < 0.001, e–g Binary LOPOCV models showing improved classification when only two classes are included in the model. Number of samples: ovarian cancer n = 39, benign n = 22, borderline tumour n = 21. Only OC samples with at least 50% of viable tumour were included

Fig. 4

Validation of the frozen tissue model using fresh tissue. a 2D PCA model of frozen ovarian cancer samples (>50% tumour content) and all normal samples combined as ‘Normal Tissue’. b 3D LDA component analysis. c Leave-one-patient-out cross-validation of the binary ‘Ovarian Cancer’ vs ‘Normal Tissue’ frozen model, showing 100% correct classification. d OMB Recognition software example showing reported classification for individual sampling points (burns) and the probability value associated with the reported class. e Bar chart showing the numbers of sampling points in fresh tissue classified with >75% probability using the OMB recognition software, plus the diagnostic accuracy at those sampling points [OC 100% (61/61), normal ovary 100% (64/64), fallopian tube 100% (58/58), peritoneum 94.1% (32/34)]

Fig. 5

Spatial resolution of metastatic peritoneal deposits and iKnife recognition. a Peritoneum containing metastatic tumour deposits with iKnife burns labelled 1–16 sampling normal peritoneum and tumour nodules. b Matched haematoxylin and eosin histological slide for the tissue sample in panel A with corresponding burn sites numbered. Scale showing millimetres. c Total ion chromatogram obtained during the sampling of the specimen with the iKnife with each burn numbered. Coloured boxes represent data obtained from a nodule (red; burns 1&2) and normal peritoneum (green; burns 9&10). d Representative mass spectra, obtained in negative-ion mode from a nodule (red) and normal peritoneum (green) showing the degree of variability in the MS peaks for different tissue types (OC vs normal). e Surgeon’s histological impression, iKnife’s impression (percentages in parenthesis represent probability of correct classification) and the histopathologist’s final diagnosis (percentages in parenthesis represent tumour cell content)