NBNathan Blake

Nathan Blake

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Nathan Blake has a total of 35 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Stainless Steel vs Calcium Fluoride for biomedical Raman Spectroscopy

    by Cynthia Ng, Nathan Blake, Geraint Thomas

    Raman spectroscopy (RS) has been shown to differentiate various disease states in tissue samples, yet the impact of sample substrate on spectral quality and downstream machine-learning (ML) performance is under-explored. We collected 960 spectra from 36 formalin-fixed, paraffin-embedded colonic specimens mounted in parallel on calcium fluoride ( CaF 2 ) and mirrored stainless-steel ( SS ) slides. Raw SS spectra were more intense than CaF 2 , and routine dewaxing left prominent paraffin peaks on CaF 2 that were largely absent on SS . Extending CaF 2 xylene immersion to 48 h reduced key paraffin peaks by 11–23 %, aligning its biochemical profile with SS . Nested cross-validated classifiers were built to distinguish true disease (T1) from epithelial misplacement (EM). The highest accuracies were observed on SS (0.65–0.66), while CaF 2 improved from 0.41–0.57 to 0.65–0.69 after extended dewaxing, chiefly via gains in recall. Although optimal dewaxing renders CaF 2 diagnostically non-inferior, SS still offers superior accuracy, faster preparation, at approximately 4% of the cost, supporting its integration into scalable, AI-enabled RS pathology workflows.

  • Open Access

    Causal responsibility based explainable AI for vibrational spectroscopy applied to oral FTIR and oesophageal Raman diagnostics

    by Nathan Blake, David A. Kelly, Sarah Kapllani-Mucaj, Rong Wang, Geraint Thomas, Hana Chockler

    Biomedical applications of vibrational spectroscopy increasingly use deep learning, but these models often give classifications without revealing which spectral features drive their decisions. We introduce Spec-ReX, a causal responsibility-based explainable AI method for vibrational spectroscopy, and compare it with the widely used explainability methods SHAP and Grad-CAM across synthetic spectra, biochemical mixtures, oral FTIR tissue spectra and oesophageal Raman tissue spectra. In a definitive in silico experiment with known discriminative peaks, Spec-ReX achieved the highest ground truth localisation, with an Intersection over Union of 0.10 and attribution concentration of 0.56, compared with 0.09 and 0.26 for the next best method. In an ambiguous in silico experiment, SHAP achieved the highest positive hit rate, while Spec-ReX was most robust to removal of low-importance regions, consistent with a sensitivity–specificity trade-off. In the in vitro biochemical experiment, Spec-ReX was again the most robust to removal of low-importance regions, while SHAP performed best when removing high-importance regions. In the ex vivo datasets, where no definitive attribution ground truth was available, the methods returned different and architecture-dependent attribution patterns. Overall, Spec-ReX provides sparse and specific model-causal responsibility maps for spectral classifiers. These explanations pertain to trained model behaviour rather than biological causality, and their clinical utility remains to be validated through user-centred studies.

  • Open Access

    Causal Responsibility Based Explainable AI for Vibrational Spectroscopy Applied to Cancer Diagnostics

    by Nathan Blake, David Kelly, Sarah Kapllani-Mucaj, Rong Wang, Geraint Thomas, Hana Chockler

    Biomedical applications of vibrational spectroscopy are benefiting from the development of deep learning. However, these models are inscrutable, with classifications given without any explanation. This ‘black-box’ problem is a key impediment to developing trust with clinicians. The need for transparent decision- making in any clinical application has resulted in regulatory requirements for explainability. Post-hoc Explainable AI (XAI) is one method used to achieve some transparency, of which there are many implementations. However, they are seldom assessed to the higher standards required for clinical applications. We introduce Spec-ReX, which uses the theory of actual causality to find features in a spectrum that are most responsible for a given classification. We rigorously assess its performance against other commonly used XAI methods on datasets of increasing complexity: in silico, in vitro and ex vivo, facilitating a rigorous and nuanced assessment.

  • AI Nurse Specialists: Transforming the nursing workforce for the age of artificial intelligence

    by Siobhan O'COnnor, Mengying Zhang, Daniel Wicks, Xiaoyang Li, Helen Caton-Peters, Peter White, Nathan Blake


  • SpecReX: Explainable AI for Raman Spectroscopy

    by Nathan Blake, David A. Kelly, Akchunya Chanchal, Sarah Kapllani-Mucaj, Geraint Thomas, Hana Chockler


  • Open Access

    I Am Big, You Are Little; I Am Right, You Are Wrong

    by David A. Kelly, Akchunya Chanchal, Nathan Blake

    Machine learning for image classification is an active and rapidly developing field. With the proliferation of classifiers of different sizes and different architectures, the problem of choosing the right model becomes more and more important. While we can assess a model's classification accuracy statistically, our understanding of the way these models work is unfortunately limited. In order to gain insight into the decision-making process of different vision models, we propose using minimal sufficient pixels sets to gauge a model's `concentration': the pixels that capture the essence of an image through the lens of the model. By comparing position, overlap, and size of sets of pixels, we identify that different architectures have statistically different concentration, in both size and position. In particular, ConvNext and EVA models differ markedly from the others. We also identify that images which are misclassified are associated with larger pixels sets than correct classifications.

  • Open Access

    Explainable AI for the Classification of Brain MRIs

    by Nathan Blake, David Kelly, Santiago Peña, Akchunya Chanchal, Hana Chockler

    Background: Machine learning applied to medical imaging labor under a lack of trust due to the inscrutability of AI models and the lack of explanations for their outcomes. Explainable AI has therefore become an essential research topic. Unfortunately, many AI models, both research and commercial, are unavailable for white-box examination, in which access to a model’s internals is required. There is therefore a need for black-box explainability tools in the medical domain. Several such tools for general images exist, but their relative strengths and weak- nesses when applied to medical images have not been explored. Methods: We use a publicly available dataset of brain MRI images and a model trained to classify cancerous and non-cancerous slices to assess a number of black-box explainability tools (LIME, RISE, IG, SHAP and ReX) and one white-box tool (Grad-CAM) as a baseline comparator. We use several common measures to assess the con- condance of the explanations with clinician provided annotations, including the Dice Coefficient, Hausdorff Distance, Jaccard Index and propose a Penalised Dice Coefficient which combines the strengths of these measures. Results: ReX (Dice Coefficient = 0.42±0.20) consistently performs relatively well across all measures with comparable performance to Grad-CAM (Dice Coef- ficient = 0.33±0.22). A panel of images is presented for qualitative inspection, showing a number of failure modes. Conclusion: In contrast to general images, we find evidence that most black-box explainability tools do not perform well for medical image classifications when used with default settings.

  • Open Access

    System transferability of Raman-based oesophageal tissue classification using modern machine learning to support multi-centre clinical diagnostics

    by Nathan Blake, Riana Gaifulina, Martin Isabelle, Jennifer Dorney, Manuel Rodriguez-Justo, Katherine Lau, Stéphanie Ohrel, Gavin Lloyd, Neil Shepherd, Aaran Lewis, et al.

    Background: The clinical potential of Raman spectroscopy is well established but has yet to become established in routine oncology workflows. One barrier slowing clinical adoption is a lack of evidence demonstrating that data taken on one spectrometer transfers across to data taken on another spectrometer to provide consistent diagnoses. Methods: We investigated multi-centre transferability using human oesophageal tissue. Raman spectra were taken across three different centres with different spectrometers of the same make and model. By using a common protocol, we aimed to minimise the difference in machine learning performance between centres. Results: 61 oesophageal samples from 51 patients were interrogated by Raman spectroscopy at each centre and classified into one of five pathologies. The overall accuracy and log-loss did not significantly vary when a model trained upon data from any one centre was applied to data taken at the other centres. Computational methods to correct for the data during pre-processing were not needed. Conclusion: We have found that when using the same make and model of spectrometer, together with a common protocol, across different centres it is possible to achieve system transferability without the need for additional computational instrument correction.


  • Open Access

    Deep Learning Applied to Raman Spectroscopy for the Detection of Microsatellite Instability/MMR Deficient Colorectal Cancer

    by Nathan Blake, Riana Gaifulina, Lewis D. Griffin, Ian M. Bell, Manuel Rodriguez-Justo, Geraint M. H. Thomas

    Defective DNA mismatch repair is one pathogenic pathway to colorectal cancer. It is characterised by microsatellite instability which provides a molecular biomarker for its detection. Clinical guidelines for universal testing of this biomarker are not met due to resource limitations; thus, there is interest in developing novel methods for its detection. Raman spectroscopy (RS) is an analytical tool able to interrogate the molecular vibrations of a sample to provide a unique biochemical fingerprint. The resulting datasets are complex and high-dimensional, making them an ideal candidate for deep learning, though this may be limited by small sample sizes. This study investigates the potential of using RS to distinguish between normal, microsatellite stable (MSS) and microsatellite unstable (MSI-H) adenocarcinoma in human colorectal samples and whether deep learning provides any benefit to this end over traditional machine learning models. A 1D convolutional neural network (CNN) was developed to discriminate between healthy, MSI-H and MSS in human tissue and compared to a principal component analysis–linear discriminant analysis (PCA–LDA) and a support vector machine (SVM) model. A nested cross-validation strategy was used to train 30 samples, 10 from each group, with a total of 1490 Raman spectra. The CNN achieved a sensitivity and specificity of 83% and 45% compared to PCA–LDA, which achieved a sensitivity and specificity of 82% and 51%, respectively. These are competitive with existing guidelines, despite the low sample size, speaking to the molecular discriminative power of RS combined with deep learning. A number of biochemical antecedents responsible for this discrimination are also explored, with Raman peaks associated with nucleic acids and collagen being implicated.


  • Open Access

    Machine Learning of Raman Spectroscopy Data for Classifying Cancers: A Review of the Recent Literature

    by Nathan Blake, Riana Gaifulina, Lewis D. Griffin, Ian M. Bell, Geraint M. H. Thomas

    Raman Spectroscopy has long been anticipated to augment clinical decision making, such as classifying oncological samples. Unfortunately, the complexity of Raman data has thus far inhibited their routine use in clinical settings. Traditional machine learning models have been used to help exploit this information, but recent advances in deep learning have the potential to improve the field. However, there are a number of potential pitfalls with both traditional and deep learning models. We conduct a literature review to ascertain the recent machine learning methods used to classify cancers using Raman spectral data. We find that while deep learning models are popular, and ostensibly outperform traditional learning models, there are many methodological considerations which may be leading to an over-estimation of performance; primarily, small sample sizes which compound sub-optimal choices regarding sampling and validation strategies. Amongst several recommendations is a call to collate large benchmark Raman datasets, similar to those that have helped transform digital pathology, which researchers can use to develop and refine deep learning models.