Relevance vector machine for optical diagnosis of cancer

Artificial Intelligence in Head and Neck Surgery

This article explores artificial intelligence's (AI's) role in otolaryngology for head and neck cancer diagnosis and management. It highlights AI's potential in pattern recognition for early cancer detection, prognostication, and treatment planning, primarily through image analysis using clinical, endoscopic, and histopathologic images. Radiomics is also discussed at length, as well as the many ways that radiologic image analysis can be utilized, including for diagnosis, lymph node metastasis prediction, and evaluation of treatment response. The study highlights AI's promise and limitations, underlining the need for clinician-data scientist collaboration to enhance head and neck cancer care.

Bibliometric analysis of the global scientific production on machine learning applied to different cancer types

Cancer disease is one of the main causes of death in the world, with million annual cases in the last decades. The need to find a cure has stimulated the search for efficient treatments and diagnostic procedures. One of the most promising tools that has emerged against cancer in recent years is machine learning (ML), which has raised a huge number of scientific papers published in a relatively short period of time. The present study analyzes global scientific production on ML applied to the most relevant cancer types through various bibliometric indicators. We find that over 30,000 studies have been published so far and observe that cancers with the highest number of published studies using ML (breast, lung, and colon cancer) are those with the highest incidence, being the USA and China the main scientific producers on the subject. Interestingly, the role of China and Japan in stomach cancer is correlated with the number of cases of this cancer type in Asia (78% of the worldwide cases). Knowing the countries and institutions that most study each area can be of great help for improving international collaborations between research groups and countries. Our analysis shows that medical and computer science journals lead the number of publications on the subject and could be useful for researchers in the field. Finally, keyword co-occurrence analysis suggests that ML-cancer research trends are focused not only on the use of ML as an effective diagnostic method, but also for the improvement of radiotherapy- and chemotherapy-based treatments.

Construction of machine learning-based models for cancer outcomes in low and lower-middle income countries: A scoping review

Background

The impact and utility of machine learning (ML)-based prediction tools for cancer outcomes including assistive diagnosis, risk stratification, and adjunctive decision-making have been largely described and realized in the high income and upper-middle-income countries. However, statistical projections have estimated higher cancer incidence and mortality risks in low and lower-middle-income countries (LLMICs). Therefore, this review aimed to evaluate the utilization, model construction methods, and degree of implementation of ML-based models for cancer outcomes in LLMICs.

Methods

PubMed/Medline, Scopus, and Web of Science databases were searched and articles describing the use of ML-based models for cancer among local populations in LLMICs between 2002 and 2022 were included. A total of 140 articles from 22,516 citations that met the eligibility criteria were included in this study.

Results

ML-based models from LLMICs were often based on traditional ML algorithms than deep or deep hybrid learning. We found that the construction of ML-based models was skewed to particular LLMICs such as India, Iran, Pakistan, and Egypt with a paucity of applications in sub-Saharan Africa. Moreover, models for breast, head and neck, and brain cancer outcomes were frequently explored. Many models were deemed suboptimal according to the Prediction model Risk of Bias Assessment tool (PROBAST) due to sample size constraints and technical flaws in ML modeling even though their performance accuracy ranged from 0.65 to 1.00. While the development and internal validation were described for all models included (n=137), only 4.4% (6/137) have been validated in independent cohorts and 0.7% (1/137) have been assessed for clinical impact and efficacy.

Conclusion

Overall, the application of ML for modeling cancer outcomes in LLMICs is increasing. However, model development is largely unsatisfactory. We recommend model retraining using larger sample sizes, intensified external validation practices, and increased impact assessment studies using randomized controlled trial designs.

Systematic review registration

https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=308345, identifier CRD42022308345.

Machine learning in point-of-care automated classification of oral potentially malignant and malignant disorders: a systematic review and meta-analysis

Machine learning (ML) algorithms are becoming increasingly pervasive in the domains of medical diagnostics and prognostication, afforded by complex deep learning architectures that overcome the limitations of manual feature extraction. In this systematic review and meta-analysis, we provide an update on current progress of ML algorithms in point-of-care (POC) automated diagnostic classification systems for lesions of the oral cavity. Studies reporting performance metrics on ML algorithms used in automatic classification of oral regions of interest were identified and screened by 2 independent reviewers from 4 databases. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were followed. 35 studies were suitable for qualitative synthesis, and 31 for quantitative analysis. Outcomes were assessed using a bivariate random-effects model following an assessment of bias and heterogeneity. 4 distinct methodologies were identified for POC diagnosis: (1) clinical photography; (2) optical imaging; (3) thermal imaging; (4) analysis of volatile organic compounds. Estimated AUROC across all studies was 0.935, and no difference in performance was identified between methodologies. We discuss the various classical and modern approaches to ML employed within identified studies, and highlight issues that will need to be addressed for implementation of automated classification systems in screening and early detection.

Accuracy of artificial intelligence-assisted detection of Oral Squamous Cell Carcinoma: A systematic review and meta-analysis

Oral Squamous Cell Carcinoma (OSCC) is an aggressive tumor with a poor prognosis. Accurate and timely diagnosis is therefore essential for reducing the burden of advanced disease and improving outcomes. In this meta-analysis, we evaluated the accuracy of artificial intelligence (AI)-assisted technologies in detecting OSCC. We included studies that validated any diagnostic modality that used AI to detect OSCC. A search was performed in six databases: PubMed, Embase, Scopus, Cochrane Library, ProQuest, and Web of Science up to 15 Mar 2022. The Quality Assessment Tool for Diagnostic Accuracy Studies was used to evaluate the included studies' quality, while the Split Component Synthesis method was utilized to quantitatively synthesize the pooled diagnostic efficacy estimates. We considered 16 out of the 566 yielded studies, which included twelve different AI models with a total of 6606 samples. The summary sensitivity, summary specificity, positive and negative likelihood ratios as well as the pooled diagnostic odds ratio were 92.0 % (95 % confidence interval [CI] 86.7-95.4 %), 91.9 % (95 % CI 86.5-95.3 %), 11.4 (95 % CI 6.74-19.2), 0.087 (95 % CI 0.051-0.146) and 132 (95 % CI 62.6-277), respectively. Our findings support the capability of AI-assisted systems to detect OSCC with high accuracy, potentially aiding the histopathological examination in early diagnosis, yet more prospective studies are needed to justify their use in the real population.

Machine-Learning Applications in Oral Cancer: A Systematic Review

Over the years, several machine-learning applications have been suggested to assist in various clinical scenarios relevant to oral cancer. We offer a systematic review to identify, assess, and summarize the evidence for reported uses in the areas of oral cancer detection and prevention, prognosis, pre-cancer, treatment, and quality of life. The main algorithms applied in the context of oral cancer applications corresponded to SVM, ANN, and LR, comprising 87.71% of the total published articles in the field. Genomic, histopathological, image, medical/clinical, spectral, and speech data were used most often to predict the four areas of application found in this review. In conclusion, our study has shown that machine-learning applications are useful for prognosis, diagnosis, and prevention of potentially malignant oral lesions (pre-cancer) and therapy. Nevertheless, we strongly recommended the application of these methods in daily clinical practice.

Role of Artificial Intelligence in the Early Diagnosis of Oral Cancer. A Scoping Review

The early diagnosis of cancer can facilitate subsequent clinical patient management. Artificial intelligence (AI) has been found to be promising for improving the diagnostic process. The aim of the present study is to increase the evidence on the application of AI to the early diagnosis of oral cancer through a scoping review. A search was performed in the PubMed, Web of Science, Embase and Google Scholar databases during the period from January 2000 to December 2020, referring to the early non-invasive diagnosis of oral cancer based on AI applied to screening. Only accessible full-text articles were considered. Thirty-six studies were included on the early detection of oral cancer based on images (photographs (optical imaging and enhancement technology) and cytology) with the application of AI models. These studies were characterized by their heterogeneous nature. Each publication involved a different algorithm with potential training data bias and few comparative data for AI interpretation. Artificial intelligence may play an important role in precisely predicting the development of oral cancer, though several methodological issues need to be addressed in parallel to the advances in AI techniques, in order to allow large-scale transfer of the latter to population-based detection protocols.

Machine learning in oral squamous cell carcinoma: Current status, clinical concerns and prospects for future-A systematic review

BACKGROUND

Oral cancer can show heterogenous patterns of behavior. For proper and effective management of oral cancer, early diagnosis and accurate prediction of prognosis are important. To achieve this, artificial intelligence (AI) or its subfield, machine learning, has been touted for its potential to revolutionize cancer management through improved diagnostic precision and prediction of outcomes. Yet, to date, it has made only few contributions to actual medical practice or patient care.

OBJECTIVES

This study provides a systematic review of diagnostic and prognostic application of machine learning in oral squamous cell carcinoma (OSCC) and also highlights some of the limitations and concerns of clinicians towards the implementation of machine learning-based models for daily clinical practice.

DATA SOURCES

We searched OvidMedline, PubMed, Scopus, Web of Science, and Institute of Electrical and Electronics Engineers (IEEE) databases from inception until February 2020 for articles that used machine learning for diagnostic or prognostic purposes of OSCC.

ELIGIBILITY CRITERIA

Only original studies that examined the application of machine learning models for prognostic and/or diagnostic purposes were considered.

DATA EXTRACTION

Independent extraction of articles was done by two researchers (A.R. & O.Y) using predefine study selection criteria. We used the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) in the searching and screening processes. We also used Prediction model Risk of Bias Assessment Tool (PROBAST) for assessing the risk of bias (ROB) and quality of included studies.

RESULTS

A total of 41 studies were published to have used machine learning to aid in the diagnosis/or prognosis of OSCC. The majority of these studies used the support vector machine (SVM) and artificial neural network (ANN) algorithms as machine learning techniques. Their specificity ranged from 0.57 to 1.00, sensitivity from 0.70 to 1.00, and accuracy from 63.4 % to 100.0 % in these studies. The main limitations and concerns can be grouped as either the challenges inherent to the science of machine learning or relating to the clinical implementations.

CONCLUSION

Machine learning models have been reported to show promising performances for diagnostic and prognostic analyses in studies of oral cancer. These models should be developed to further enhance explainability, interpretability, and externally validated for generalizability in order to be safely integrated into daily clinical practices. Also, regulatory frameworks for the adoption of these models in clinical practices are necessary.

Fluorescence photo-bleaching of urine and its applicability in oral cancer diagnosis

Photo-stability of urine is of crucial importance for the applicability of fluorescence spectroscopy of urine samples for diagnosis of cancer. We report the results of a detailed study on fluorescence photo-bleaching of human urine samples. We also present the results of a preliminary investigation on evaluation of the applicability of photo-bleaching characteristics of urine for discriminating patients with oral cancer from healthy volunteers. The time-lapse fluorescence induced by continuous shining of 405 nm radiation from a diode laser was recorded from the urine samples obtained from 18 patients with oral cancer as well as from 22 healthy volunteers with history of no known major illness in the past two months. The integrated fluorescence intensity (ΣI), calculated for each spectrum, was found to decrease with time till a point after which no further decrease was observed. Further, while significant differences were observed in the spectra of cancerous patients and healthy volunteers, these differences were found to be varying with time till the intensities of the observed fluorescence spectra corresponding to the two categories of urine samples became stable. The curve, generated by plotting ΣI vs. time, was found to be best fitted (R² > 0.95) with a double-exponential decay function. The photo-bleaching constants, obtained from curve-fitting, were found to have statistically significant differences corresponding to the urine samples of cancerous patients and healthy volunteers. A classification algorithm developed based on nearest-mean classifier (NMC) and applied on the photo-bleaching constants in leave-one-subject-out cross-validation mode was found to provide a sensitivity and specificity of up to ∼ 86% in discriminating the two categories of urine samples.

Automatic Recognition of Ripening Tomatoes by Combining Multi-Feature Fusion with a Bi-Layer Classification Strategy for Harvesting Robots

Automatic recognition of ripening tomatoes is a main hurdle precluding the replacement of manual labour by robotic harvesting. In this paper, we present a novel automatic algorithm for recognition of ripening tomatoes using an improved method that combines multiple features, feature analysis and selection, a weighted relevance vector machine (RVM) classifier, and a bi-layer classification strategy. The algorithm operates using a two-layer strategy. The first-layer classification strategy aims to identify tomato-containing regions in images using the colour difference information. The second classification strategy is based on a classifier that is trained on multi-medium features. In our proposed algorithm, to simplify the calculation and to improve the recognition efficiency, the processed images are divided into 9 × 9 pixel blocks, and these blocks, rather than single pixels, are considered as the basic units in the classification task. Six colour-related features, namely the Red (R), Green (G), Blue (B), Hue (H), Saturation (S) and Intensity (I) components, respectively, colour components, and five textural features (entropy, energy, correlation, inertial moment and local smoothing) were extracted from pixel blocks. Relevant features and their weights were analysed using the iterative RELIEF (I-RELIEF) algorithm. The image blocks were classified into different categories using a weighted RVM classifier based on the selected relevant features. The final results of tomato recognition were determined by combining the block classification results and the bi-layer classification strategy. The algorithm demonstrated the detection accuracy of 94.90% on 120 images, this suggests that the proposed algorithm is effective and suitable for tomato detection

Visualization and tissue classification of human breast cancer images using ultrahigh-resolution OCT

BACKGROUND AND OBJECTIVE

Breast cancer is one of the most common cancers, and recognized as the third leading cause of mortality in women. Optical coherence tomography (OCT) enables three dimensional visualization of biological tissue with micrometer level resolution at high speed, and can play an important role in early diagnosis and treatment guidance of breast cancer. In particular, ultra-high resolution (UHR) OCT provides images with better histological correlation. This paper compared UHR OCT performance with standard OCT in breast cancer imaging qualitatively and quantitatively. Automatic tissue classification algorithms were used to automatically detect invasive ductal carcinoma in ex vivo human breast tissue.

STUDY DESIGN/MATERIALS AND METHODS

Human breast tissues, including non-neoplastic/normal tissues from breast reduction and tumor samples from mastectomy specimens, were excised from patients at Columbia University Medical Center. The tissue specimens were imaged by two spectral domain OCT systems at different wavelengths: a home-built ultra-high resolution (UHR) OCT system at 800 nm (measured as 2.72 μm axial and 5.52 μm lateral) and a commercial OCT system at 1,300 nm with standard resolution (measured as 6.5 μm axial and 15 μm lateral), and their imaging performances were analyzed qualitatively. Using regional features derived from OCT images produced by the two systems, we developed an automated classification algorithm based on relevance vector machine (RVM) to differentiate hollow-structured adipose tissue against solid tissue. We further developed B-scan based features for RVM to classify invasive ductal carcinoma (IDC) against normal fibrous stroma tissue among OCT datasets produced by the two systems. For adipose classification, 32 UHR OCT B-scans from 9 normal specimens, and 28 standard OCT B-scans from 6 normal and 4 IDC specimens were employed. For IDC classification, 152 UHR OCT B-scans from 6 normal and 13 IDC specimens, and 104 standard OCT B-scans from 5 normal and 8 IDC specimens were employed.

RESULTS

We have demonstrated that UHR OCT images can produce images with better feature delineation compared with images produced by 1,300 nm OCT system. UHR OCT images of a variety of tissue types found in human breast tissue were presented. With a limited number of datasets, we showed that both OCT systems can achieve a good accuracy in identifying adipose tissue. Classification in UHR OCT images achieved higher sensitivity (94%) and specificity (93%) of adipose tissue than the sensitivity (91%) and specificity (76%) in 1,300 nm OCT images. In IDC classification, similarly, we achieved better results with UHR OCT images, featured an overall accuracy of 84%, sensitivity of 89% and specificity of 71% in this preliminary study.

CONCLUSION

In this study, we provided UHR OCT images of different normal and malignant breast tissue types, and qualitatively and quantitatively studied the texture and optical features from OCT images of human breast tissue at different resolutions. We developed an automated approach to differentiate adipose tissue, fibrous stroma, and IDC within human breast tissues. Our work may open the door toward automatic intraoperative OCT evaluation of early-stage breast cancer. Lasers Surg. Med. 49:258-269, 2017. © 2017 Wiley Periodicals, Inc.

Accuracy of autofluorescence in diagnosing oral squamous cell carcinoma and oral potentially malignant disorders: a comparative study with aero-digestive lesions

Presently, various studies had investigated the accuracy of autofluorescence in diagnosing oral squamous cell carcinoma (OSCC) and oral potentially malignant disorders (OPMD) with diverse conclusions. This study aimed to assess its accuracy for OSCC and OPMD and to investigate its applicability in general dental practice. After a comprehensive literature search, a meta-analysis was conducted to calculate the pooled diagnostic indexes of autofluorescence for premalignant lesions (PML) and malignant lesions (ML) of the oral cavity, lung, esophagus, stomach and colorectum and to compute indexes regarding the detection of OSCC aided by algorithms. Besides, a u test was performed. Twenty-four studies detecting OSCC and OPMD in 2761 lesions were included. This demonstrated that the overall accuracy of autofluorescence for OSCC and OPMD was superior to PML and ML of the lung, esophagus and stomach, slightly inferior to the colorectum. Additionally, the sensitivity and specificity for OSCC and OPMD were 0.89 and 0.8, respectively. Furthermore, the specificity could be remarkably improved by additional algorithms. With relatively high accuracy, autofluorescence could be potentially applied as an adjunct for early diagnosis of OSCC and OPMD. Moreover, approaches such as algorithms could enhance its specificity to ensure its efficacy in primary care.

In vivo Raman spectroscopy for detection of oral neoplasia: a pilot clinical study

We report a pilot study carried out to evaluate the applicability of in vivo Raman spectroscopy for differential diagnosis of malignant and potentially malignant lesions of human oral cavity in a clinical setting. The study involved 28 healthy volunteers and 171 patients having various lesions of oral cavity. The Raman spectra, measured from multiple sites of normal oral mucosa and of lesions belonging to three histopathological categories, viz. oral squamous cell carcinoma (OSCC), oral submucous fibrosis (OSMF) and leukoplakia (OLK), were subjected to a probability based multivariate statistical algorithm capable of direct multi-class classification. With respect to histology as the gold standard, the diagnostic algorithm was found to provide an accuracy of 85%, 89%, 85% and 82% in classifying the oral tissue spectra into the four tissue categories based on leave-one-subject-out cross validation. When employed for binary classification, the algorithm resulted in a sensitivity and specificity of 94% in discriminating normal from the rest of the abnormal spectra of OSCC, OSMF and OLK tissue sites pooled together.

Integrated Device for in Vivo Fine Needle Aspiration Biopsy and Elastic Scattering Spectroscopy in Preoperative Thyroid Nodules

Thyroid nodules are a frequent clinical finding and the most common endocrine malignancy is thyroid cancer. The standard of care in the management of a patient with a thyroid nodule is to perform a preoperative fine needle aspiration (FNA) biopsy of the suspect nodule under ultrasound imaging guidance. In a significant percentage of the cases, cytological assessment of the biopsy material yields indeterminate results, the consequence of which is diagnostic thyroidectomy. Unfortunately, 75–80% of diagnostic thyroidectomies following indeterminate cytology result in benign designation by post-surgery histopathology, indicating potentially unnecessary surgeries. Clearly, the potential exists for the improvement in patient care and the reduction of overall procedure costs if an improved preoperative diagnostic technique was developed. Elastic scattering spectroscopy (ESS) is an optical biopsy technique that is mediated by optical fiber probes and has been shown to be effective in differentiating benign from malignant thyroid tissue in ex vivo surgical tissue samples. The goal of the current research was to integrate the ESS fiber optic probes into a device that can also collect cells for cytological assessment and, thus, enable concurrent spectroscopic interrogation and biopsy of a suspect nodule with a single needle penetration. The primary challenges to designing the device included miniaturizing the standard ESS fiber optic probe to fit within an FNA needle and maintaining the needle’s aspiration functionality. We demonstrate the value of the fabricated prototype devices by assessing their preliminary performance in an on-going clinical study with >120 patients. The devices have proven to be clinically friendly, collecting both aspirated cells and optical data from the same location in thyroid nodules and with minimal disruption of clinical procedure. In the future, such integrated devices could be used to complement FNA-based cytological results and have the potential to both reduce the number of diagnostic thyroidectomies on benign nodules and improve the surgical approach for patients with thyroid malignancies, thereby, decreasing healthcare costs and improving patient outcomes.

Improved modeling of clinical data with kernel methods

OBJECTIVE

Despite the rise of high-throughput technologies, clinical data such as age, gender and medical history guide clinical management for most diseases and examinations. To improve clinical management, available patient information should be fully exploited. This requires appropriate modeling of relevant parameters.

METHODS

When kernel methods are used, traditional kernel functions such as the linear kernel are often applied to the set of clinical parameters. These kernel functions, however, have their disadvantages due to the specific characteristics of clinical data, being a mix of variable types with each variable its own range. We propose a new kernel function specifically adapted to the characteristics of clinical data.

RESULTS

The clinical kernel function provides a better representation of patients' similarity by equalizing the influence of all variables and taking into account the range r of the variables. Moreover, it is robust with respect to changes in r. Incorporated in a least squares support vector machine, the new kernel function results in significantly improved diagnosis, prognosis and prediction of therapy response. This is illustrated on four clinical data sets within gynecology, with an average increase in test area under the ROC curve (AUC) of 0.023, 0.021, 0.122 and 0.019, respectively. Moreover, when combining clinical parameters and expression data in three case studies on breast cancer, results improved overall with use of the new kernel function and when considering both data types in a weighted fashion, with a larger weight assigned to the clinical parameters. The increase in AUC with respect to a standard kernel function and/or unweighted data combination was maximum 0.127, 0.042 and 0.118 for the three case studies.

CONCLUSION

For clinical data consisting of variables of different types, the proposed kernel function--which takes into account the type and range of each variable--has shown to be a better alternative for linear and non-linear classification problems.

Spectral classifier design with ensemble classifiers and misclassification-rejection: application to elastic-scattering spectroscopy for detection of colonic neoplasia

Optical spectroscopy has shown potential as a real-time, in vivo, diagnostic tool for identifying neoplasia during endoscopy. We present the development of a diagnostic algorithm to classify elastic-scattering spectroscopy (ESS) spectra as either neoplastic or non-neoplastic. The algorithm is based on pattern recognition methods, including ensemble classifiers, in which members of the ensemble are trained on different regions of the ESS spectrum, and misclassification-rejection, where the algorithm identifies and refrains from classifying samples that are at higher risk of being misclassified. These "rejected" samples can be reexamined by simply repositioning the probe to obtain additional optical readings or ultimately by sending the polyp for histopathological assessment, as per standard practice. Prospective validation using separate training and testing sets result in a baseline performance of sensitivity = .83, specificity = .79, using the standard framework of feature extraction (principal component analysis) followed by classification (with linear support vector machines). With the developed algorithm, performance improves to Se ∼ 0.90, Sp ∼ 0.90, at a cost of rejecting 20-33% of the samples. These results are on par with a panel of expert pathologists. For colonoscopic prevention of colorectal cancer, our system could reduce biopsy risk and cost, obviate retrieval of non-neoplastic polyps, decrease procedure time, and improve assessment of cancer risk.

INTEGRATED OPTICAL TOOLS FOR MINIMALLY INVASIVE DIAGNOSIS AND TREATMENT AT GASTROINTESTINAL ENDOSCOPY

Over the past two decades, the bulk of gastrointestinal (GI) endoscopic procedures has shifted away from diagnostic and therapeutic interventions for symptomatic disease toward cancer prevention in asymptomatic patients. This shift has resulted largely from a decrease in the incidence of peptic ulcer disease in the era of antisecretory medications coupled with emerging evidence for the efficacy of endoscopic detection and eradication of dysplasia, a histopathological biomarker widely accepted as a precursor to cancer. This shift has been accompanied by a drive toward minimally-invasive, in situ optical diagnostic technologies that help assess the mucosa for cellular changes that relate to dysplasia. Two competing but complementary approaches have been pursued. The first approach is based on broad-view targeting of "areas of interest" or "red flags." These broad-view technologies include standard white light endoscopy (WLE), high-definition endoscopy (HD), and "electronic" chromoendoscopy (narrow-band-type imaging). The second approach is based on multiple small area or point-source (meso/micro) measurements, which can be either machine (spectroscopy) or human-interpreted (endomicroscopy, magnification endoscopy), much as histopatholgy slides are. In this paper we present our experience with the development and testing of a set of familiar but "smarter" standard tissue-sampling tools that can be routinely employed during screening/surveillance endoscopy. These tools have been designed to incorporate fiberoptic probes that can mediate spectroscopy or endomicroscopy. We demonstrate the value of such tools by assessing their preliminary performance from several ongoing clinical studies. Our results have shown promise for a new generation of integrated optical tools for a variety of screening/surveillance applications during GI endoscopy. Integrated devices should prove invaluable for dysplasia surveillance strategies that currently result in large numbers of benign biopsies, which are of little clinical consequence, including screening for colorectal polyps and surveillance of "flat" dysplasia such as Barrett's esophagus and chronic colitis due to inflammatory bowel diseases.

Cost-sensitive case-based reasoning using a genetic algorithm: application to medical diagnosis

OBJECTIVE

The paper studies the new learning technique called cost-sensitive case-based reasoning (CSCBR) incorporating unequal misclassification cost into CBR model. Conventional CBR is now considered as a suitable technique for diagnosis, prognosis and prescription in medicine. However it lacks the ability to reflect asymmetric misclassification and often assumes that the cost of a positive diagnosis (an illness) as a negative one (no illness) is the same with that of the opposite situation. Thus, the objective of this research is to overcome the limitation of conventional CBR and encourage applying CBR to many real world medical cases associated with costs of asymmetric misclassification errors.

METHODS

The main idea involves adjusting the optimal cut-off classification point for classifying the absence or presence of diseases and the cut-off distance point for selecting optimal neighbors within search spaces based on similarity distribution. These steps are dynamically adapted to new target cases using a genetic algorithm. We apply this proposed method to five real medical datasets and compare the results with two other cost-sensitive learning methods-C5.0 and CART.

RESULTS

Our finding shows that the total misclassification cost of CSCBR is lower than other cost-sensitive methods in many cases. Even though the genetic algorithm has limitations in terms of unstable results and over-fitting training data, CSCBR results with GA are better overall than those of other methods. Also the paired t-test results indicate that the total misclassification cost of CSCBR is significantly less than C5.0 and CART for several datasets.

CONCLUSION

We have proposed a new CBR method called cost-sensitive case-based reasoning (CSCBR) that can incorporate unequal misclassification costs into CBR and optimize the number of neighbors dynamically using a genetic algorithm. It is meaningful not only for introducing the concept of cost-sensitive learning to CBR, but also for encouraging the use of CBR in the medical area. The result shows that the total misclassification costs of CSCBR do not increase in arithmetic progression as the cost of false absence increases arithmetically, thus it is cost-sensitive. We also show that total misclassification costs of CSCBR are the lowest among all methods in four datasets out of five and the result is statistically significant in many cases. The limitation of our proposed CSCBR is confined to classify binary cases for minimizing misclassification cost because our proposed CSCBR is originally designed to classify binary case. Our future work extends this method for multi-classification which can classify more than two groups.

Development of a kernel function for clinical data

For most diseases and examinations, clinical data such as age, gender and medical history guides clinical management, despite the rise of high-throughput technologies. To fully exploit such clinical information, appropriate modeling of relevant parameters is required. As the widely used linear kernel function has several disadvantages when applied to clinical data, we propose a new kernel function specifically developed for this data. This "clinical kernel function" more accurately represents similarities between patients. Evidently, three data sets were studied and significantly better performances were obtained with a Least Squares Support Vector Machine when based on the clinical kernel function compared to the linear kernel function.

Preoperative diagnosis of ovarian tumors using Bayesian kernel-based methods

OBJECTIVES

To develop flexible classifiers that predict malignancy in adnexal masses using a large database from nine centers.

METHODS

The database consisted of 1066 patients with at least one persistent adnexal mass for which a large amount of clinical and ultrasound data were recorded. The outcome of interest was the histological classification of the adnexal mass as benign or malignant. The outcome was predicted using Bayesian least squares support vector machines in comparison with relevance vector machines. The models were developed on a training set (n=754) and tested on a test set (n=312).

RESULTS

Twenty-five percent of the patients (n=266) had a malignant tumor. Variable selection resulted in a set of 12 variables for the models: age, maximal diameter of the ovary, maximal diameter of the solid component, personal history of ovarian cancer, hormonal therapy, very strong intratumoral blood flow (i.e. color score 4), ascites, presumed ovarian origin of tumor, multilocular-solid tumor, blood flow within papillary projections, irregular internal cyst wall and acoustic shadows. Test set area under the receiver-operating characteristics curve (AUC) for all models exceeded 0.940, with a sensitivity above 90% and a specificity above 80% for all models. The least squares support vector machine model with linear kernel performed very well, with an AUC of 0.946, 91% sensitivity and 84% specificity. The models performed well in the test sets of all the centers.

CONCLUSIONS

Bayesian kernel-based methods can accurately separate malignant from benign masses. The robustness of the models will be investigated in future studies.

A probability-based spectroscopic diagnostic algorithm for simultaneous discrimination of brain tumor and tumor margins from normal brain tissue

This paper reports the development of a probability-based spectroscopic diagnostic algorithm capable of simultaneously discriminating tumor core and tumor margins from normal human brain tissues. The algorithm uses a nonlinear method for feature extraction based on maximum representation and discrimination feature (MRDF) and a Bayesian method for classification based on sparse multinomial logistic regression (SMLR). Both the autofluorescence and the diffuse-reflectance spectra acquired in vivo from patients undergoing craniotomy or temporal lobectomy at the Vanderbilt University Medical Center were used to train and validate the algorithm. The classification accuracy was observed to be approximately 96%, 80%, and 97% for the tumor, tumor margin, and normal brain tissues, respectively, for the training data set and approximately 96%, 94%, and 100%, respectively, for the corresponding tissue types in an independent validation data set. The inherently multi-class nature of the algorithm facilitates a rapid and simultaneous classification of tissue spectra into various tissue categories without the need for a hierarchical multi-step binary classification scheme. Further, the probabilistic nature of the algorithm makes it possible to quantitatively assess the certainty of the classification and recheck the samples that are classified with higher relative uncertainty.

Multi-class classification algorithm for optical diagnosis of oral cancer

We report development of a direct multi-class spectroscopic diagnostic algorithm for discrimination of high-grade cancerous tissue sites from low-grade as well as precancerous and normal squamous tissue sites of human oral cavity. The algorithm was developed making use of the recently formulated theory of total principal component regression (TPCR). The in vivo autofluorescence spectral data acquired from patients screened for neoplasm of oral cavity at the Government Cancer Hospital, Indore, was used to train and validate the algorithm. The diagnostic algorithm based on TPCR was found to provide satisfactory performance in classifying the tissue sites in four different classes - high-grade squamous cell carcinoma, low-grade squamous cell carcinoma, leukoplakia, and normal squamous tissue. The classification accuracy for these four classes was observed to be approximately 94%, 100%, 100% and 91% for the training data set (based on leave-one-out cross-validation), and was approximately 90%, 90%, 85% and 88%, respectively for the corresponding classes for the independent validation data set.

N2 laser excited autofluorescence spectroscopy of formalin-fixed human breast tissue

The paper reports results of an in vitro study on autofluorescence spectroscopy of fresh and formalin-fixed human breast tissue samples to investigate the effect of formalin fixation on the measured autofluorescence spectra. It also explores the applicability of the approach in discriminating cancerous from the uninvolved sites of the formalin-fixed breast tissues based on their autofluorescence spectra. A probability-based diagnostic algorithm, making use of the theory of relevance vector machine (RVM), a powerful recent approach for statistical pattern recognition, was developed for that purpose. The algorithm provided sensitivity values of up to 97% and specificity values of up to 100% towards cancer for both the independent validation data set as well as for the training data set based on leave-one-out cross-validation. These results suggest that autofluorescence spectroscopy may prove to be a valuable additional in vitro diagnostic modality in clinical pathology setting for discriminating cancerous tissue sites from normal sites.