The Right Direction Needed to Develop White-Box Deep Learning in Radiology, Pathology, and Ophthalmology: A Short Review

A refined deep-learning-based algorithm for harmful-algal-bloom remote-sensing recognition using Noctiluca scintillans algal bloom as an example

Harmful algal blooms (HABs) are challenging to recognize because of their striped and uneven biomass distributions. To address this issue, a refined deep-learning algorithm termed HAB-Ne was developed for the recognition of HABs in GF-1 Wide Field of View (WFV) images using Noctiluca scintillans algal bloom as an example. First, a pretrained image super-resolution model was integrated to improve the spatial resolution of the GF-1 WFV images and minimize the impact of mixed pixels caused by the strip distribution. Side-window convolution was also explored to enhance the edge features of HABs and minimize the effects of uneven biomass distribution. In addition, a convolutional encoder-decoder network was constructed for threshold-free HAB recognition to address the dependence on thresholds in existing methods. HAB-Net effectively recognized HABs from GF-1 WFV images, achieving an average precision of 90.1% and an F1-score of 0.86. HAB-Net showed more fine-grained recognition results than those of existing methods, with over 4% improvement in the F1-Score, especially in the marginal areas of HAB distribution. The algorithm demonstrated its effectiveness in recognizing HABs in different marine environments, such as the Yellow Sea, East China Sea, and northern Vietnam. Additionally, the algorithm was proven suitable for detecting the macroalga Sargassum. This study demonstrates the potential of deep-learning-based fine-grained recognition of HABs, which can be extended to the recognition of other fine-scale and strip-distributed objects, such as oil spills and Ulva prolifera.

Cancer immunotherapy efficacy and machine learning

INTRODUCTION

Immunotherapy is one of the major breakthroughs in the treatment of cancer, and it has become a powerful clinical strategy, however, not all patients respond to immune checkpoint blockade and other immunotherapy strategies. Applying machine learning (ML) techniques to predict the efficacy of cancer immunotherapy is useful for clinical decision-making.

AREAS COVERED

Applying ML including deep learning (DL) in radiomics, pathomics, tumor microenvironment (TME) and immune-related genes analysis to predict immunotherapy efficacy. The studies in this review were searched from PubMed and ClinicalTrials.gov (January 2023).

EXPERT OPINION

An increasing number of studies indicate that ML has been applied to various aspects of oncology research, with the potential to provide more effective individualized immunotherapy strategies and enhance treatment decisions. With advances in ML technology, more efficient methods of predicting the efficacy of immunotherapy may become available in the future.

Classification of tumor types using XGBoost machine learning model: a vector space transformation of genomic alterations

BACKGROUND

Machine learning (ML) represents a powerful tool to capture relationships between molecular alterations and cancer types and to extract biological information. Here, we developed a plain ML model aimed at distinguishing cancer types based on genetic lesions, providing an additional tool to improve cancer diagnosis, particularly for tumors of unknown origin.

METHODS

TCGA data from 9,927 samples spanning 32 different cancer types were downloaded from cBioportal. A vector space model type data transformation technique was designed to build consistently homogeneous new datasets containing, as predictive features, calls for somatic point mutations and copy number variations at chromosome arm-level, thus allowing the use of the XGBoost classifier models. Considering the imbalance in the dataset, due to large difference in the number of cases for each tumor, two preprocessing strategies were considered: i) setting a percentage cut-off threshold to remove less represented cancer types, ii) dividing cancer types into different groups based on biological criteria and training a specific XGBoost model for each of them. The performance of all trained models was mainly assessed by the out-of-sample balanced accuracy (BACC) and the AUC scores.

RESULTS

The XGBoost classifier achieved the best performance (BACC 77%; AUC 97%) on a dataset containing the 10 most represented tumor types. Moreover, dividing the 18 most represented cancers into three different groups (endocrine-related carcinomas, other carcinomas and other cancers),such analysis models achieved 78%, 71% and 86% BACC, respectively, with AUC scores greater than 96%. In addition, the model capable of linking each group to a specific cancer type reached 81% BACC and 94% AUC. Overall, the diagnostic potential of our model was comparable/higher with respect to others already described in literature and based on similar molecular data and ML approaches.

CONCLUSIONS

A boosted ML approach able to accurately discriminate different cancer types was developed. The methodology builds datasets simpler and more interpretable than the original data, while keeping enough information to accurately train standard ML models without resorting to sophisticated Deep Learning architectures. In combination with histopathological examinations, this approach could improve cancer diagnosis by using specific DNA alterations, processed by a replicable and easy-to-use automated technology. The study encourages new investigations which could further increase the classifier's performance, for example by considering more features and dividing tumors into their main molecular subtypes.

Systematic review: fluid biomarkers and machine learning methods to improve the diagnosis from mild cognitive impairment to Alzheimer's disease

Mild cognitive impairment (MCI) is often considered an early stage of dementia, with estimated rates of progression to dementia up to 80-90% after approximately 6 years from the initial diagnosis. Diagnosis of cognitive impairment in dementia is typically based on clinical evaluation, neuropsychological assessments, cerebrospinal fluid (CSF) biomarkers, and neuroimaging. The main goal of diagnosing MCI is to determine its cause, particularly whether it is due to Alzheimer's disease (AD). However, only a limited percentage of the population has access to etiological confirmation, which has led to the emergence of peripheral fluid biomarkers as a diagnostic tool for dementias, including MCI due to AD. Recent advances in biofluid assays have enabled the use of sophisticated statistical models and multimodal machine learning (ML) algorithms for the diagnosis of MCI based on fluid biomarkers from CSF, peripheral blood, and saliva, among others. This approach has shown promise for identifying specific causes of MCI, including AD. After a PRISMA analysis, 29 articles revealed a trend towards using multimodal algorithms that incorporate additional biomarkers such as neuroimaging, neuropsychological tests, and genetic information. Particularly, neuroimaging is commonly used in conjunction with fluid biomarkers for both cross-sectional and longitudinal studies. Our systematic review suggests that cost-effective longitudinal multimodal monitoring data, representative of diverse cultural populations and utilizing white-box ML algorithms, could be a valuable contribution to the development of diagnostic models for AD due to MCI. Clinical assessment and biomarkers, together with ML techniques, could prove pivotal in improving diagnostic tools for MCI due to AD.

Machine Learning for Automated Classification of Abnormal Lung Sounds Obtained from Public Databases: A Systematic Review

Pulmonary auscultation is essential for detecting abnormal lung sounds during physical assessments, but its reliability depends on the operator. Machine learning (ML) models offer an alternative by automatically classifying lung sounds. ML models require substantial data, and public databases aim to address this limitation. This systematic review compares characteristics, diagnostic accuracy, concerns, and data sources of existing models in the literature. Papers published from five major databases between 1990 and 2022 were assessed. Quality assessment was accomplished with a modified QUADAS-2 tool. The review encompassed 62 studies utilizing ML models and public-access databases for lung sound classification. Artificial neural networks (ANN) and support vector machines (SVM) were frequently employed in the ML classifiers. The accuracy ranged from 49.43% to 100% for discriminating abnormal sound types and 69.40% to 99.62% for disease class classification. Seventeen public databases were identified, with the ICBHI 2017 database being the most used (66%). The majority of studies exhibited a high risk of bias and concerns related to patient selection and reference standards. Summarizing, ML models can effectively classify abnormal lung sounds using publicly available data sources. Nevertheless, inconsistent reporting and methodologies pose limitations to advancing the field, and therefore, public databases should adhere to standardized recording and labeling procedures.

Automated analysis of vessel morphometry in retinal images from a Danish high street optician setting

PURPOSE

To evaluate the test performance of the QUARTZ (QUantitative Analysis of Retinal vessel Topology and siZe) software in detecting retinal features from retinal images captured by health care professionals in a Danish high street optician chain, compared with test performance from other large population studies (i.e., UK Biobank) where retinal images were captured by non-experts.

METHOD

The dataset FOREVERP (Finding Ophthalmic Risk and Evaluating the Value of Eye exams and their predictive Reliability, Pilot) contains retinal images obtained from a Danish high street optician chain. The QUARTZ algorithm utilizes both image processing and machine learning methods to determine retinal image quality, vessel segmentation, vessel width, vessel classification (arterioles or venules), and optic disc localization. Outcomes were evaluated by metrics including sensitivity, specificity, and accuracy and compared to human expert ground truths.

RESULTS

QUARTZ's performance was evaluated on a subset of 3,682 images from the FOREVERP database. 80.55% of the FOREVERP images were labelled as being of adequate quality compared to 71.53% of UK Biobank images, with a vessel segmentation sensitivity of 74.64% and specificity of 98.41% (FOREVERP) compared with a sensitivity of 69.12% and specificity of 98.88% (UK Biobank). The mean (± standard deviation) vessel width of the ground truth was 16.21 (4.73) pixels compared to that predicted by QUARTZ of 17.01 (4.49) pixels, resulting in a difference of -0.8 (1.96) pixels. The differences were stable across a range of vessels. The detection rate for optic disc localisation was similar for the two datasets.

CONCLUSION

QUARTZ showed high performance when evaluated on the FOREVERP dataset, and demonstrated robustness across datasets, providing validity to direct comparisons and pooling of retinal feature measures across data sources.

Application of urinary biomarkers for diagnosing acute kidney injury in critically ill patients without baseline renal function data

PURPOSE

Estimating the baseline renal function of patients without prior creatinine measurement is crucial for diagnosing acute kidney injury (AKI). This study aimed to incorporate AKI biomarkers into a new AKI diagnosis rule when no premorbid baseline is available.

METHODS

This prospective observational study was conducted in an adult intensive care unit (ICU). Urinary neutrophil gelatinase-associated lipocalin (NGAL) and L-type fatty acid-binding protein (L-FABP) were measured at ICU admission. An AKI diagnostic rule was composed by classification and regression tree (CART) analysis.

RESULTS

A total of 243 patients were enrolled. In the development cohort, CART analysis composed a decision tree for AKI diagnosis selecting serum creatinine and urinary NGAL at ICU admission as predictors. In the validation cohort, the novel decision rule was superior to the imputation strategy based on Modification of Diet in Renal Disease (MDRD) equation regarding misclassification rate (13.0% vs. 29.6%, p = 0.002). Decision curve analysis demonstrated that the net benefit of the decision rule exceeded the MDRD approach in a threshold probability range of 25% and higher.

CONCLUSIONS

The novel diagnostic rule incorporating serum creatinine and urinary NGAL at ICU admission showed superiority to the MDRD approach in AKI diagnosis without baseline renal function data.

Progress on deep learning in digital pathology of breast cancer: a narrative review

BACKGROUND AND OBJECTIVE

Pathology is the gold standard criteria for breast cancer diagnosis and has important guiding value in formulating the clinical treatment plan and predicting the prognosis. However, traditional microscopic examinations of tissue sections are time consuming and labor intensive, with unavoidable subjective variations. Deep learning (DL) can evaluate and extract the most important information from images with less need for human instruction, providing a promising approach to assist in the pathological diagnosis of breast cancer. To provide an informative and up-to-date summary on the topic of DL-based diagnostic systems for breast cancer pathology image analysis and discuss the advantages and challenges to the routine clinical application of digital pathology.

METHODS

A PubMed search with keywords ("breast neoplasm" or "breast cancer") and ("pathology" or "histopathology") and ("artificial intelligence" or "deep learning") was conducted. Relevant publications in English published from January 2000 to October 2021 were screened manually for their title, abstract, and even full text to determine their true relevance. References from the searched articles and other supplementary articles were also studied.

KEY CONTENT AND FINDINGS

DL-based computerized image analysis has obtained impressive achievements in breast cancer pathology diagnosis, classification, grading, staging, and prognostic prediction, providing powerful methods for faster, more reproducible, and more precise diagnoses. However, all artificial intelligence (AI)-assisted pathology diagnostic models are still in the experimental stage. Improving their economic efficiency and clinical adaptability are still required to be developed as the focus of further researches.

CONCLUSIONS

Having searched PubMed and other databases and summarized the application of DL-based AI models in breast cancer pathology, we conclude that DL is undoubtedly a promising tool for assisting pathologists in routines, but further studies are needed to realize the digitization and automation of clinical pathology.

Machine-Learning-Based Disease Diagnosis: A Comprehensive Review

Globally, there is a substantial unmet need to diagnose various diseases effectively. The complexity of the different disease mechanisms and underlying symptoms of the patient population presents massive challenges in developing the early diagnosis tool and effective treatment. Machine learning (ML), an area of artificial intelligence (AI), enables researchers, physicians, and patients to solve some of these issues. Based on relevant research, this review explains how machine learning (ML) is being used to help in the early identification of numerous diseases. Initially, a bibliometric analysis of the publication is carried out using data from the Scopus and Web of Science (WOS) databases. The bibliometric study of 1216 publications was undertaken to determine the most prolific authors, nations, organizations, and most cited articles. The review then summarizes the most recent trends and approaches in machine-learning-based disease diagnosis (MLBDD), considering the following factors: algorithm, disease types, data type, application, and evaluation metrics. Finally, in this paper, we highlight key results and provides insight into future trends and opportunities in the MLBDD area.

Applications of interpretability in deep learning models for ophthalmology

PURPOSE OF REVIEW

In this article, we introduce the concept of model interpretability, review its applications in deep learning models for clinical ophthalmology, and discuss its role in the integration of artificial intelligence in healthcare.

RECENT FINDINGS

The advent of deep learning in medicine has introduced models with remarkable accuracy. However, the inherent complexity of these models undermines its users' ability to understand, debug and ultimately trust them in clinical practice. Novel methods are being increasingly explored to improve models' 'interpretability' and draw clearer associations between their outputs and features in the input dataset. In the field of ophthalmology, interpretability methods have enabled users to make informed adjustments, identify clinically relevant imaging patterns, and predict outcomes in deep learning models.

SUMMARY

Interpretability methods support the transparency necessary to implement, operate and modify complex deep learning models. These benefits are becoming increasingly demonstrated in models for clinical ophthalmology. As quality standards for deep learning models used in healthcare continue to evolve, interpretability methods may prove influential in their path to regulatory approval and acceptance in clinical practice.

Machine learning-based decision tree classifier for the diagnosis of progressive supranuclear palsy and corticobasal degeneration

Aims

This study aimed to clarify the different topographical distribution of tau pathology between progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD) and establish a machine learning‐based decision tree classifier.

Methods

Paraffin‐embedded sections of the temporal cortex, motor cortex, caudate nucleus, globus pallidus, subthalamic nucleus, substantia nigra, red nucleus, and midbrain tectum from 1020 PSP and 199 CBD cases were assessed by phospho‐tau immunohistochemistry. The severity of tau lesions (i.e., neurofibrillary tangle, coiled body, tufted astrocyte or astrocytic plaque, and tau threads) was semi‐quantitatively scored in each region. Hierarchical cluster analysis was performed using tau pathology scores. A decision tree classifier was made with tau pathology scores using 914 cases. Cross‐validation was done using 305 cases. An additional ten cases were used for a validation study.

Results

Cluster analysis displayed two distinct clusters; the first cluster included only CBD, and the other cluster included all PSP and six CBD cases. We built a decision tree, which used only seven decision nodes. The scores of tau threads in the caudate nucleus were the most decisive factor for predicting CBD. In a cross‐validation, 302 out of 305 cases were correctly diagnosed. In the pilot validation study, three investigators made a correct diagnosis in all cases using the decision tree.

Conclusion

Regardless of the morphology of astrocytic tau lesions, semi‐quantitative tau pathology scores in select brain regions are sufficient to distinguish PSP and CBD. The decision tree simplifies neuropathologic differential diagnosis of PSP and CBD.

Improving prediction for medical institution with limited patient data: Leveraging hospital-specific data based on multicenter collaborative research network

BACKGROUND AND OBJECTIVE

Clinical decision support assisted by prediction models usually faces the challenges of limited clinical data and a lack of labels when the model is developed with data from a single medical institution. Accordingly, research on multicenter clinical collaborative networks, which can provide external medical data, has received increasing attention. With the increasing availability of machine learning techniques such as transfer learning, leveraging large-scale patient data from multiple hospitals to build data-driven predictive models with clinical application potential provides an alternative solution to address the problem of limited patient data.

METHODS

A multicenter hybrid semi-supervised transfer learning model (MHSTL) is proposed in this study on the basis of unified common data model to ensure multicenter data standardized representation. Then the hospital-specific features, along with the co-occurrence features across domains, are aligned through a representation learning architecture that is built based on deep neural networks and the newly proposed neural decision forest model. In this process, limited patient data from the target hospital, both labeled and unlabeled, are incorporated during the feature adaptation process, thereby contributing to better model performance. Without patient-level data sharing, the proposed model learning strategy which overcomes feature misalignment and distribution divergence, enables the multi-source transfer learning process in the case of insufficient and unlabeled patient data at target hospital.

RESULTS

The effectiveness of the proposed transfer learning model was evaluated on a collaborative research network of colorectal cancer patients in the US and China. The results demonstrate that the proposed model can achieve much better performance for predicting target risk with limited resources on patient data than baseline models      . Better discrimination and calibration ability are also observed when sufficient labeled data are not available in the target hospital for prognosis prediction tasks      . Further exploratory experiments show that the proposed approach exhibits good model generalizability regardless of the data heterogeneity. With the help of the SHapley Additive exPlanations for model interpretation, the effectiveness of incorporating hospital-specific features in the transfer learning model is shown.

CONCLUSIONS

In this study, the proposed method can develop prediction models from multiple source hospitals and exhibit good performance by leveraging cross-domain hospital-specific feature information, therefore enhancing the model prediction when applied to single medical institution with limited patient data.

One-Dimensional Convolutional Neural Networks with Feature Selection for Highly Concise Rule Extraction from Credit Scoring Datasets with Heterogeneous Attributes

Convolution neural networks (CNNs) have proven effectiveness, but they are not applicable to all datasets, such as those with heterogeneous attributes, which are often used in the finance and banking industries. Such datasets are difficult to classify, and to date, existing high-accuracy classifiers and rule-extraction methods have not been able to achieve sufficiently high classification accuracies or concise classification rules. This study aims to provide a new approach for achieving transparency and conciseness in credit scoring datasets with heterogeneous attributes by using a one-dimensional (1D) fully-connected layer first CNN combined with the Recursive-Rule Extraction (Re-RX) algorithm with a J48graft decision tree (hereafter 1D FCLF-CNN). Based on a comparison between the proposed 1D FCLF-CNN and existing rule extraction methods, our architecture enabled the extraction of the most concise rules (6.2) and achieved the best accuracy (73.10%), i.e., the highest interpretability–priority rule extraction. These results suggest that the 1D FCLF-CNN with Re-RX with J48graft is very effective for extracting highly concise rules for heterogeneous credit scoring datasets. Although it does not completely overcome the accuracy–interpretability dilemma for deep learning, it does appear to resolve this issue for credit scoring datasets with heterogeneous attributes, and thus, could lead to a new era in the financial industry.

Bayesian Proxy Modelling for Estimating Black Carbon Concentrations using White-Box and Black-Box Models

Black carbon (BC) is an important component of particulate matter (PM) in urban environments. BC is typically emitted from gas and diesel engines, coal-fired power plants, and other sources that burn fossil fuel. In contrast to PM, BC measurements are not always available on a large scale due to the operational cost and complexity of the instrumentation. Therefore, it is advantageous to develop a mathematical model for estimating the quantity of BC in the air, termed a BC proxy, to enable widening of spatial air pollution mapping. This article presents the development of BC proxies based on a Bayesian framework using measurements of PM concentrations and size distributions from 10 to 10,000 nm from a recent mobile air pollution study across several areas of Jordan. Bayesian methods using informative priors can naturally prevent over-fitting in the modelling process and the methods generate a confidence interval around the prediction, thus the estimated BC concentration can be directly quantified and assessed. In particular, two types of models are developed based on their transparency and interpretability, referred to as white-box and black-box models. The proposed methods are tested on extensive data sets obtained from the measurement campaign in Jordan. In this study, black-box models perform slightly better due to their model complexity. Nevertheless, the results demonstrate that the performance of both models does not differ significantly. In practice, white-box models are relatively more convenient to be deployed, the methods are well understood by scientists, and the models can be used to better understand key relationships.