pith. sign in

arxiv: 2605.05729 · v1 · submitted 2026-05-07 · 📡 eess.SP

Optimizing In Vivo Oral Lesion Classification from Electrical Impedance Spectroscopy Using Data-driven Approaches

Sophie A. Lloyd , Jacob P. Th\"ones , Safina S. Suratwala , Noor Zaghlula , Liang Lu , Joseph Paydarfar , Ethan K. Murphy , Sascha Spors
show 1 more author
Ryan J. Halter
This is my paper

Pith reviewed 2026-05-08 07:26 UTC · model grok-4.3

classification 📡 eess.SP
keywords electrical impedance spectroscopyoral cancermachine learningdimensionality reductionlesion classificationPCAin vivo diagnosticsbinary classification
0
0 comments X

The pith

Machine learning reduces electrical impedance data by 99% for accurate oral lesion classification.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper develops a pipeline that applies principal component analysis and pattern selection to electrical impedance spectroscopy measurements from oral tissues. It demonstrates that using far fewer data points from a handheld device can classify lesions as healthy, cancerous, or other types more accurately than models using the complete dataset. A logistic regression model reached 80 percent accuracy and 0.90 area under the curve for distinguishing healthy from cancerous tissue in patient-group validation. Such efficiency supports practical bedside use for early cancer detection where full data processing would be cumbersome.

Core claim

By ranking frequencies and reducing current injection and voltage measurement patterns, the input dimensionality for classifying oral pathology from in vivo EIS data can be cut by up to 99 percent while improving diagnostic accuracy over baseline full-dataset models, with logistic regression achieving 80 percent accuracy and an AUC of 0.90 for binary healthy-versus-cancer tasks and AUCs above 0.82 for multi-class scenarios.

What carries the argument

The central mechanism is independent frequency ranking combined with PCA dimensionality reduction and testing of reduced IIVV pattern geometries, validated via leave-one-patient-group-out cross-validation on data from 104 patients.

If this is right

  • All top-performing models relied on the significantly reduced IIVV set as input.
  • The approach yields computationally efficient classifiers suitable for real-time clinical application on handheld devices.
  • Multi-class discrimination maintains AUCs above 0.82 even after the large reduction in input size.
  • The pipeline methodology is readily generalizable to other EIS devices and applications.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This reduction could enable deployment on low-power portable devices for point-of-care screening in underserved regions.
  • Similar data-driven optimization might improve impedance-based diagnostics for other tissue types or diseases.
  • Extending the method to incorporate additional patient metadata could further boost performance without increasing data volume.

Load-bearing premise

That the leave-one-patient-group-out cross-validation on measurements from 104 patients produces models that generalize reliably to new patients and that the selected reduced patterns retain all information needed for accurate clinical decisions.

What would settle it

Observing substantially lower classification accuracy or AUC when applying the reduced-pattern models to an independent set of new patients compared to the full-data baselines would indicate the reduction discards essential information or overfits to the study cohort.

Figures

Figures reproduced from arXiv: 2605.05729 by Ethan K. Murphy, Jacob P. Th\"ones, Joseph Paydarfar, Liang Lu, Noor Zaghlula, Ryan J. Halter, Safina S. Suratwala, Sascha Spors, Sophie A. Lloyd.

Figure 1
Figure 1. Figure 1: PCB electrode array with numbered electrodes. The center contains view at source ↗
Figure 2
Figure 2. Figure 2: Optimization schedule for reducing the amount of EIS data sample view at source ↗
Figure 3
Figure 3. Figure 3: The impedance magnitude of all four tissue types across frequencies. view at source ↗
Figure 5
Figure 5. Figure 5: Reduced Frequency 95 % Confidence Interval of Mean AUCs for All Tasks. CI shown by error bar, black outline on colored markers designate statistically higher AUC as compared to all frequencies (p < 0.05). reduce the number of parameters: geometry-based IIVV sets, and impedance threshold (z-threshold) set. Both utilize the full frequency spectrum for each evaluation. The z-threshold that resulted in the hig… view at source ↗
Figure 4
Figure 4. Figure 4: Task 1 classification AUCs using all IIVVs and reduced frequencies, with AUC as the training metric. Each line shows the averaged AUC from ten trials of five-fold cross-validation. Results from each fold of the first of ten trials are shown in view at source ↗
Figure 6
Figure 6. Figure 6: IIVV selection results for all tasks, tuned with AUC scoring. The view at source ↗
Figure 7
Figure 7. Figure 7: Per-class and micro-average ROC performance for view at source ↗
Figure 8
Figure 8. Figure 8: Timings of the full clinical and computational workflow for one measurement. Data collection, with one measurement with 3 bursts recorded, filtering view at source ↗
read the original abstract

Oral cancer is a significant global health burden, and early detection remains a critical clinical need. Electrical impedance spectroscopy (EIS) offers a promising non-invasive approach for real-time tissue characterization, but classification frameworks that jointly leverage multiple impedance features for in vivo oral lesion discrimination remain underdeveloped. This paper presents a machine-learning (ML) pipeline to optimize classification of in vivo oral pathology from EIS data collected using a handheld, bedside device. Impedance measurements were acquired from 104 patients undergoing oral cancer resection or biopsy. Three classification tasks were evaluated: (1) healthy vs. cancer, (2) multi-class lesion-type discrimination (cancer, high-grade dysplasia, non-malignant), and (3) multi-class discrimination between the three lesion pathologies and healthy tissue. For each task, signal frequencies were independently ranked and reduced using PCA, and different current injection/voltage measurement (IIVV) pattern geometries were tested. Classification performance was assessed through leave-one-patient-group-out cross-validation to ensure robustness on unseen patients. Input data dimensionality was reduced by up to 99% across all tasks while improving diagnostic accuracy over baseline models trained on the full dataset. A logistic regression model achieved the highest binary classification accuracy of 80% with an AUC of 0.90, while multi-class scenarios maintained AUCs above 0.82. All top-performing models utilized the significantly reduced IIVV set as input. The proposed pipeline advances EIS-based cancer detection by providing a robust, computationally efficient, and clinically practical framework for early diagnosis of oral cancer lesions, with a methodology readily generalizable to other EIS devices and applications.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents a data-driven machine-learning pipeline for classifying in vivo oral lesions from electrical impedance spectroscopy (EIS) measurements collected with a handheld device on 104 patients. For three tasks (healthy vs. cancer; multi-class lesion types; and lesion types plus healthy), frequencies are ranked and reduced via PCA, multiple current-injection/voltage-measurement (IIVV) pattern geometries are tested, and performance is evaluated with leave-one-patient-group-out cross-validation. The central claims are that input dimensionality can be reduced by up to 99 % while improving accuracy over full-dataset baselines, that logistic regression reaches 80 % accuracy and AUC 0.90 on the binary task, and that multi-class AUCs remain above 0.82, with all top models using the reduced IIVV sets.

Significance. If the performance gains are free of selection bias, the work would offer a practical, low-dimensional, non-invasive framework for real-time oral-cancer detection that is computationally efficient and potentially generalizable to other EIS devices. The patient-group CV design is a positive step toward assessing generalization to unseen patients.

major comments (2)
  1. [Abstract / Methods (CV procedure)] Abstract and cross-validation description: the manuscript states that frequencies were independently ranked, PCA was applied, and IIVV geometries were tested, with performance assessed via leave-one-patient-group-out CV. It does not indicate whether ranking, PCA-component selection, or IIVV-geometry choice were performed inside each training fold (nested CV) or on the full dataset before CV. If the latter occurred, the 99 % dimensionality reduction and the reported accuracy improvements over baseline are optimistically biased, directly undermining the claim that the reduced patterns preserve all clinically relevant information without selection bias. This is load-bearing for the central empirical results.
  2. [Results] Results section: no quantitative baseline accuracies or AUCs for the full-dataset models are reported, no confidence intervals or error bars accompany the 80 % / 0.90 and >0.82 figures, and no details are given on the frequency-ranking criterion or the exact number of PCA components retained. These omissions prevent assessment of the magnitude and statistical reliability of the claimed improvements.
minor comments (2)
  1. [Abstract] The abstract notes class imbalance as a potential concern but provides no description of how it was handled (e.g., class weights, oversampling, or stratified folds).
  2. Missing references to prior EIS oral-lesion studies or standard ML baselines (e.g., random forest, SVM) would help situate the logistic-regression result.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their insightful and constructive comments, which have prompted us to clarify important methodological aspects of our study. We provide detailed responses to each major comment below, along with our plans for revision.

read point-by-point responses

  1. Referee: [Abstract / Methods (CV procedure)] Abstract and cross-validation description: the manuscript states that frequencies were independently ranked, PCA was applied, and IIVV geometries were tested, with performance assessed via leave-one-patient-group-out CV. It does not indicate whether ranking, PCA-component selection, or IIVV-geometry choice were performed inside each training fold (nested CV) or on the full dataset before CV. If the latter occurred, the 99 % dimensionality reduction and the reported accuracy improvements over baseline are optimistically biased, directly undermining the claim that the reduced patterns preserve all clinically relevant information without selection bias. This is load-bearing for the central empirical results.

    Authors: We acknowledge the referee's concern regarding potential selection bias. The frequency ranking, PCA, and IIVV selection were performed on the full dataset to determine the reduced configurations that would be used in practice. While this is a common approach in exploratory studies, we agree it can inflate performance estimates. Therefore, we will revise the manuscript to employ nested cross-validation: an inner loop for feature selection and PCA within each outer leave-one-patient-group-out fold. This will provide a more reliable assessment of the 99% reduction's effectiveness. The abstract and methods will be updated to describe this procedure, and we will re-evaluate the performance metrics accordingly. We believe this addresses the load-bearing issue for our central claims. revision: yes

  2. Referee: [Results] Results section: no quantitative baseline accuracies or AUCs for the full-dataset models are reported, no confidence intervals or error bars accompany the 80 % / 0.90 and >0.82 figures, and no details are given on the frequency-ranking criterion or the exact number of PCA components retained. These omissions prevent assessment of the magnitude and statistical reliability of the claimed improvements.

    Authors: We agree that these details are essential for transparency and evaluation. In the revised version, we will report the baseline accuracies and AUCs for models using the complete set of frequencies and IIVV patterns. We will add 95% confidence intervals (via bootstrapping over patient groups) to all performance figures. We will also detail the frequency-ranking method (e.g., using ANOVA F-statistics for class separability) and the criterion for retaining PCA components (e.g., cumulative variance threshold of 95%). These additions will enable readers to gauge the improvements' magnitude and reliability. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical ML pipeline with held-out CV

full rationale

The paper reports an entirely data-driven classification study on EIS measurements from 104 patients. It ranks frequencies, applies PCA, tests IIVV geometries, and evaluates via leave-one-patient-group-out cross-validation, with performance compared to full-dataset baselines. No mathematical derivation chain, no self-referential predictions, no fitted parameters renamed as independent results, and no load-bearing self-citations exist. All reported accuracies and AUCs are direct empirical outcomes on unseen patient groups, satisfying the self-contained benchmark criterion.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper is an applied empirical study that relies on standard machine-learning and statistical assumptions rather than new physical postulates or derivations.

axioms (2)
  • domain assumption Leave-one-patient-group-out cross-validation produces unbiased estimates of generalization performance on unseen patients
    Invoked to ensure robustness on unseen patients as stated in the abstract.
  • domain assumption Principal component analysis on ranked frequencies preserves the clinically discriminative information in the impedance spectra
    Used to justify the 99% dimensionality reduction while claiming maintained or improved accuracy.

pith-pipeline@v0.9.0 · 5627 in / 1561 out tokens · 52942 ms · 2026-05-08T07:26:00.744447+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 2 canonical work pages

  1. [1]

    Cancer statistics, 2025,

    R. L. Siegel, T. B. Kratzer, A. N. Giaquinto, H. Sung, and A. Jemal, “Cancer statistics, 2025,”CA: A Cancer Journal for Clinicians, vol. 75, no. 1, pp. 10–45, 2025. eprint: https://acsjournals.onlinelibrary.wiley.com/doi/pdf/10.3322/caac.21871

    work page doi:10.3322/caac.21871 2025

  2. [2]

    Cancer of the oral cavity and pharynx - cancer stat facts

    N. C. Institute, “Cancer of the oral cavity and pharynx - cancer stat facts.”

  3. [3]

    The pathology of oral cancer,

    P. M. Speight and P. M. Farthing, “The pathology of oral cancer,”British Dental Journal, vol. 225, no. 9, pp. 841–847, 2018. Num Pages: 7 Place: London, United States Publisher: Nature Publishing Group

    2018

  4. [4]

    Why oral histopathology suffers inter-observer variability on grading oral epithelial dysplasia: an attempt to understand the sources of variation,

    O. Kujan, A. Khattab, R. J. Oliver, S. A. Roberts, N. Thakker, and P. Sloan, “Why oral histopathology suffers inter-observer variability on grading oral epithelial dysplasia: an attempt to understand the sources of variation,”Oral Oncol, vol. 43, no. 3, pp. 224–231, 2007

    2007

  5. [5]

    Interobserver agreement in dysplasia grading: toward an enhanced gold standard for clinical pathology trials,

    P. M. Speight, T. J. Abram, P. N. Floriano, R. James, J. Vick, M. H. Thornhill, C. Murdoch, C. Freeman, A. M. Hegarty, K. D’Apice, A. R. Kerr, J. Phelan, P. Corby, I. Khouly, N. Vigneswaran, J. Bouquot, N. M. Demian, Y . E. Weinstock, S. W. Redding, S. Rowan, C.-K. Yeh, H. S. McGuff, F. R. Miller, and J. T. McDevitt, “Interobserver agreement in dysplasia ...

    2015

  6. [6]

    Experiences, perceptions, and decision-making capacity towards oral biopsy among dental students and dentists,

    J. Cassol Spanemberg, R. Vel ´azquez Cay ´on, J. Romanini, M. A. Trevizani Martins, P. L ´opez-Jornet, and V . C. Carrard, “Experiences, perceptions, and decision-making capacity towards oral biopsy among dental students and dentists,”Sci Rep, vol. 13, no. 1, p. 22937, 2023. Publisher: Nature Publishing Group

    2023

  7. [7]

    Optical fluorescence imag- ing in oral cancer and potentially malignant disorders: A systematic review,

    L. Tiwari, O. Kujan, and C. S. Farah, “Optical fluorescence imag- ing in oral cancer and potentially malignant disorders: A systematic review,”Oral Diseases, vol. 26, no. 3, pp. 491–510, 2020. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/odi.13071

    work page doi:10.1111/odi.13071 2020

  8. [8]

    Understanding the biological basis of autofluores- cence imaging for oral cancer detection: high-resolution fluorescence microscopy in viable tissue,

    I. Pavlova, M. Williams, A. El-Naggar, R. Richards-Kortum, and A. Gillenwater, “Understanding the biological basis of autofluores- cence imaging for oral cancer detection: high-resolution fluorescence microscopy in viable tissue,”Clin Cancer Res, vol. 14, no. 8, pp. 2396– 2404, 2008

    2008

  9. [9]

    Intraop- erative use of wide-field optical coherence tomography to evaluate tissue microstructure in the oral cavity and oropharynx,

    A. K. Badhey, J. S. Schwarz, B. M. Laitman, B. M. Veremis, W. H. Westra, M. Yao, M. S. Teng, E. M. Genden, and B. A. Miles, “Intraop- erative use of wide-field optical coherence tomography to evaluate tissue microstructure in the oral cavity and oropharynx,”JAMA Otolaryngol Head Neck Surg, vol. 149, no. 1, pp. 71–78, 2023

    2023

  10. [10]

    Technology review: The use of electrical impedance scanning in the detection of breast cancer,

    T. A. Hope and S. E. Iles, “Technology review: The use of electrical impedance scanning in the detection of breast cancer,”Breast Cancer Research, vol. 6, no. 2, p. 69, 2003

    2003

  11. [11]

    Electrical impedance spectroscopy of the human prostate,

    R. J. Halter, A. Hartov, J. A. Heaney, K. D. Paulsen, and A. R. Schned, “Electrical impedance spectroscopy of the human prostate,” IEEE Transactions on Biomedical Engineering, vol. 54, no. 7, pp. 1321– 1327, 2007. Conference Name: IEEE Transactions on Biomedical Engineering

    2007

  12. [12]

    Use of electrical impedance spectroscopy to detect malig- nant and potentially malignant oral lesions,

    C. Murdoch, B. H. Brown, V . Hearnden, P. M. Speight, K. D’Apice, A. M. Hegarty, J. A. Tidy, T. J. Healey, P. E. Highfield, and M. H. Thornhill, “Use of electrical impedance spectroscopy to detect malig- nant and potentially malignant oral lesions,”International journal of nanomedicine, pp. 4521–4532, 2014

    2014

  13. [13]

    In vivo classification of oral lesions using electrical impedance spectroscopy,

    S. A. Lloyd, T. E. Lee, E. K. Murphy, A. F. Doussan, J. P. Th ¨ones, D. A. Kerr, J. A. Paydarfar, and R. J. Halter, “In vivo classification of oral lesions using electrical impedance spectroscopy,”IEEE Transactions on Biomedical Engineering, pp. 1–10, 2025

    2025

  14. [14]

    The use of bioimpedance in the detection/screening of tongue cancer,

    T.-P. Sun, C. T.-S. Ching, C.-S. Cheng, S.-H. Huang, Y .-J. Chen, C.-S. Hsiao, C.-H. Chang, S.-Y . Huang, H.-L. Shieh, W.-H. Liu, C.-M. Liu, and C.-Y . Chen, “The use of bioimpedance in the detection/screening of tongue cancer,”Cancer Epidemiol, vol. 34, no. 2, pp. 207–211, 2010

    2010

  15. [15]

    A preliminary study of the use of bioimpedance in the screening of squamous tongue cancer,

    C. T.-S. Ching, T.-P. Sun, S.-H. Huang, C.-S. Hsiao, C.-H. Chang, S.-Y . Huang, Y .-J. Chen, C.-S. Cheng, H.-L. Shieh, and C.-Y . Chen, “A preliminary study of the use of bioimpedance in the screening of squamous tongue cancer,”Int J Nanomedicine, vol. 5, pp. 213–220, 2010

    2010

  16. [16]

    Tissue level based deep learning framework for early detection of dysplasia in oral squamous epithelium,

    R. K. Gupta, M. Kaur, and J. Manhas, “Tissue level based deep learning framework for early detection of dysplasia in oral squamous epithelium,” Journal of Multimedia Information System, vol. 6, no. 2, pp. 81–86, 2019

    2019

  17. [17]

    Computer-assisted medical image classification for early diagnosis of oral cancer employing deep learning algorithm,

    P. R. Jeyaraj and E. R. Samuel Nadar, “Computer-assisted medical image classification for early diagnosis of oral cancer employing deep learning algorithm,”Journal of cancer research and clinical oncology, vol. 145, pp. 829–837, 2019

    2019

  18. [18]

    Automatic classification of cancerous tissue in laserendomicroscopy images of the oral cavity using deep learning,

    M. Aubreville, C. Knipfer, N. Oetter, C. Jaremenko, E. Rodner, J. Den- zler, C. Bohr, H. Neumann, F. Stelzle, and A. Maier, “Automatic classification of cancerous tissue in laserendomicroscopy images of the oral cavity using deep learning,”Scientific reports, vol. 7, no. 1, p. 11979, 2017

    2017

  19. [19]

    An early diagnosis of oral cancer based on three-dimensional convolutional neural networks,

    S. Xu, C. Liu, Y . Zong, S. Chen, Y . Lu, L. Yang, E. Y . K. Ng, Y . Wang, Y . Wang, Y . Liu, W. Hu, and C. Zhang, “An early diagnosis of oral cancer based on three-dimensional convolutional neural networks,”IEEE Access, vol. 7, pp. 158603–158611, 2019

    2019

  20. [20]

    Deep learning-based electrical impedance spectroscopy analysis for malignant and potentially malignant oral disorder detection,

    Z. Lin, Z.-Q. Lang, L. Guo, D. C. Walker, M. Matella, M. Wang, and C. Murdoch, “Deep learning-based electrical impedance spectroscopy analysis for malignant and potentially malignant oral disorder detection,” Sci Rep, vol. 15, no. 1, p. 19458, 2025. Publisher: Nature Publishing Group

    2025

  21. [21]

    Electrical impedance-based tissue classification for bladder tumor differentiation,

    C. Veil, F. Krauß, B. Amend, F. Fend, and O. Sawodny, “Electrical impedance-based tissue classification for bladder tumor differentiation,” Sci Rep, vol. 15, no. 1, p. 825, 2025. Publisher: Nature Publishing Group

    2025

  22. [22]

    IIVV Evaluation

    J. P. Th ¨ones, “IIVV Evaluation.” https://github.com/EITLabworks/IIVV- Evaluation, 2026

    2026

  23. [23]

    A clinically feasible electrode array for 3d intraoperative electrical impedance tomography-based surgical margin assessment in robot-assisted radical prostatectomy,

    S. E. Kossmann, E. K. Murphy, A. F. Doussan, S. Lloyd, and R. J. Halter, “A clinically feasible electrode array for 3d intraoperative electrical impedance tomography-based surgical margin assessment in robot-assisted radical prostatectomy,”IEEE Transactions on Biomedical Engineering, vol. 71, no. 11, pp. 3134–3145, 2024

    2024

  24. [24]

    Comparison of complex open domain electrical impedance tomography methods,

    A. F. Doussan, E. K. Murphy, S. A. Lloyd, and R. J. Halter, “Comparison of complex open domain electrical impedance tomography methods,” IEEE Transactions on Biomedical Engineering, pp. 1–12, 2025. 13

    2025

  25. [25]

    Binary-and three-tiered oral epithelial dysplasia grading system and malignant transformation,

    R. Ellonen, J. Kelppe, J. Hagstr ¨om, A. Suominen, J. Willberg, J. Rautava, and H. K. Laine, “Binary-and three-tiered oral epithelial dysplasia grading system and malignant transformation,”Oral Diseases, 2025

    2025

  26. [26]

    Oral epithelial dysplasia: Do we have a management solution? a systematic review,

    N. Mannapperuma, C. Y . Chieng, and V . Ilankovan, “Oral epithelial dysplasia: Do we have a management solution? a systematic review,” Advances in Oral and Maxillofacial Surgery, vol. 18, p. 100533, 2025

    2025

  27. [27]

    Scikit-learn: Machine learning in Python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vander- plas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duch- esnay, “Scikit-learn: Machine learning in Python,”Journal of Machine Learning Research, vol. 12, pp. 2825–2830, 2011

    2011

  28. [28]

    Principal component analysis,

    M. Greenacre, P. J. Groenen, T. Hastie, A. I. d’Enza, A. Markos, and E. Tuzhilina, “Principal component analysis,”Nature Reviews Methods Primers, vol. 2, no. 1, p. 100, 2022

    2022

  29. [29]

    Support-vector networks,

    C. Cortes and V . Vapnik, “Support-vector networks,”Machine learning, vol. 20, no. 3, pp. 273–297, 1995

    1995

  30. [30]

    Support vector machines,

    M. Hearst, S. Dumais, E. Osuna, J. Platt, and B. Scholkopf, “Support vector machines,”IEEE Intelligent Systems and their Applications, vol. 13, no. 4, pp. 18–28, 1998

    1998

  31. [31]

    Random forests,

    L. Breiman, “Random forests,”Machine Learning, vol. 45, pp. 5–32, Oct 2001

    2001

  32. [32]

    D. W. Hosmer, S. Lemeshow, and R. X. Sturdivant,Applied logistic regression. Wiley New York, 2000

    2000

  33. [33]

    The meaning and use of the area under a receiver operating characteristic (roc) curve.,

    J. A. Hanley and B. J. McNeil, “The meaning and use of the area under a receiver operating characteristic (roc) curve.,”Radiology, vol. 143, no. 1, pp. 29–36, 1982

    1982

  34. [34]

    Machine learning applications in cancer prognosis and prediction,

    K. Kourou, T. P. Exarchos, K. P. Exarchos, M. V . Karamouzis, and D. I. Fotiadis, “Machine learning applications in cancer prognosis and prediction,”Computational and structural biotechnology journal, vol. 13, pp. 8–17, 2015

    2015

  35. [35]

    Low-frequency dielectric properties of the oral mucosa,

    I. Lackovic and Z. Stare, “Low-frequency dielectric properties of the oral mucosa,” in13th International Conference on Electrical Bioimpedance and the 8th Conference on Electrical Impedance Tomography(H. Schar- fetter and R. Merwa, eds.), vol. 17, pp. 154–157, Springer Berlin Heidelberg, 2007. Series Title: IFMBE Proceedings

    2007