Azzam Taktak

The use of artificial neural networks in decision support in cancer: A systematic review

Artificial neural networks have featured in a wide range of medical journals, often with promising results. This paper reports on a systematic review that was conducted to assess the benefit of artificial neural networks (ANNs) as decision making tools in the field of cancer. The number of clinical trials (CTs) and randomised controlled trials (RCTs) involving the use of ANNs in diagnosis and prognosis increased from 1 to 38 in the last decade. However, out of 396 studies involving the use of ANNs in cancer, only 27 were either CTs or RCTs. Out of these trials, 21 showed an increase in benefit to healthcare provision and 6 did not. None of these studies however showed a decrease in benefit. This paper reviews the clinical fields where neural network methods figure most prominently, the main algorithms featured, methodologies for model selection and the need for rigorous evaluation of results.

Estimating prognosis for survival after treatment of choroidal melanoma

Choroidal melanoma is fatal in about 50% of patients. This is because of metastatic disease, which usually involves the liver. Kaplan-Meier survival curves based only on tumor size and extent do not give a true indication of prognosis. This is because the survival prognosis of choroidal melanoma correlates not only with clinical stage but also with histologic grade, genetic type, and competing causes of death. We have developed an online tool that predicts survival using all these data also taking normal life-expectancy into account. The estimated prognosis is accurate enough to be relevant to individual patients. Such personalized prognostication improves the well-being of patients having an excellent survival probability, not least because it spares them from unnecessary screening tests. Such screening can be targeted at high-risk patients, so that metastases are detected sooner, thereby enhancing any opportunities for treatment. Concerns about psychological harm have proved exaggerated. At least in Britain, patients want to know their prognosis, even if this is poor. The ability to select patients with a high risk of metastasis improves prospects for randomised studies evaluating systemic adjuvant therapy aimed at preventing or delaying metastatic disease. Furthermore, categorization of tissue samples according to survival prognosis enables laboratory studies to be undertaken without waiting many years for survival to be measured. As a result of advances in histologic and genetic studies, biopsy techniques and statistics, prognostication has become established as a routine procedure in our clinical practice, thereby enhancing the care of patients with uveal melanoma.

A fuzzy system for detecting distorted plethysmogram pulses in neonates and paediatric patients

Pulse oximetry is a useful, quick, non-invasive and widely used technology for monitoring oxygen saturation (SaO2) for neonates and paediatric patients. However, pulse oximetry is fraught with artefacts, causing false alarms resulting from patient or probe movement. The shape of the plethysmogram is a useful visual indicator for determining the reliability of SaO2 numerical readings. If certain features could be defined that tag valid plethysmogram pulses, then automatic recognition of valid SaO2 values can be attained. We observed that the systolic upstroke time (t1), the diastolic time (t2) and heart rate (HR) extracted from the plethysmogram pulse constitute features which can be used for detecting normal and distorted plethysmogram pulses. Therefore, we developed a knowledge-based system using fuzzy logic for classifying plethysmogram pulses into two categories: valid and artefact. A total of 22,497 pulse waveforms were used to define the system parameters. These were obtained from 13 patients with heart rates ranging between 62 and 209 beats min-1. A further 1420 waveforms obtained from another four patients were used for testing the system, and visually classified into 833 (59%) valid and 587 (41%) distorted segments. The system was able to classify 679 (82%) valid segments and 543 (93%) distorted segments correctly. The calculations of the systems performance showed 82% sensitivity, 86% accuracy and 93% specificity. We, therefore, conclude that the algorithm used in this system can be implemented in its present from for real-time SaO2 monitoring in intensive care for detecting valid and distorted plethysmogram pulses.

Automatic detection of distorted plethysmogram pulses in neonates and paediatric patients using an adaptive-network-based fuzzy inference system

Despite the fact that pulse oximetry has become an essential technology in respiratory monitoring of neonates and paediatric patients, it is still fraught with artefacts causing false alarms resulting from patient or probe movement. As the shape of the plethysmogram has always been considered as a useful visual indicator for determining the reliability of SaO(2) numerical readings, automation of this observation might benefit health care providers at the bedside. We observed that the systolic upstroke time (t(1)), the diastolic time (t(2)) and heart rate (HR) extracted from the plethysmogram pulse constitute features, which can be used for detecting normal and distorted plethysmogram pulses. We developed a technique for classifying plethysmogram pulses into two categories: valid and artefact via implementations of fuzzy inference systems (FIS), which were tuned using an adaptive-network-based fuzzy inference system (ANFIS) and receiver operating characteristics (ROC) curves analysis. Features extracted from a total of 22,497 pulse waveforms obtained from 13 patients were used to systematically optimise the FIS. A further 2843 waveforms obtained from another eight patients were used for testing the system, and visually classified into 1635 (58%) valid and 1208 (42%) distorted segments. For the optimum system, the area under the ROC curve was 0.92. The system was able to classify 1418 (87%) valid segments and 897 (74%) distorted segments correctly. The calculations of the systems performance showed 87% sensitivity, 81% accuracy and 74% specificity. In comparison with the 95% confidence interval (CI) thresholding method, the fuzzy system showed higher specificity (P=0.008,P<0.01), and no significant difference was found between the two methods in terms of sensitivity (P=0.720,P>0.05) and accuracy (P=0.053,P>0.05). We therefore conclude that the algorithm used in this system has some potential in detecting valid and distorted plethysmogram pulse. However, further evaluation is needed using larger patient groups.

Genetic Heterogeneity in Uveal Melanoma Assessed by Multiplex Ligation-Dependent Probe Amplification

PURPOSE To determine intratumor genetic heterogeneity in uveal melanoma (UM) by multiplex ligation-dependent probe amplification (MLPA) in formalin-fixed, paraffin-embedded (FFPE) tumor tissues. METHODS DNA was extracted from whole tumor sections and from two to nine different areas microdissected from 32 FFPE UMs. Thirty-one loci on chromosomes 1, 3, 6, and 8 were tested with MLPA for copy number changes. The tumor was considered heterogeneous at a locus if (1) the difference in dosage quotients (DQs) of any two areas was 0.2 or more, and (2) the DQs of the areas belonged to different ranges. RESULTS Comparison of MLPA data obtained from microdissected areas of the UMs showed heterogeneity in 1 to 26 examined loci in 24 (75%) tumors, with only 25% of the tumors being homogeneous. Intratumor heterogeneity of 3p12.2, 6p21.2, and 8q11.23 was most common, occurring in >30% of the UMs. Gains of chromosome 3 were observed in four UMs, with three of these tumors showing the highest degree of heterogeneity. Copy number variation was associated with differences in tumor cell type, but not with differences in tumor pigmentation or reactive inflammation. UMs with genetic heterogeneity across multiple sample sites showed equivocal MLPA results when the whole tumor section was examined. These results suggest that different clones dilute MLPA results. CONCLUSIONS Heterogeneity of chromosomal abnormalities of chromosomes 1, 3, 6, and 8 is present in most UMs. This heterogeneity causes equivocal MLPA results. One random tumor sample may not be representative of the whole tumor and, therefore, may be insufficient for prognostic testing.

Multiplex ligation-dependent probe amplification of conjunctival melanoma reveals common BRAF V600E gene mutation and gene copy number changes.

PURPOSE To determine the occurrence of BRAF V600E gene mutations and copy number changes of all autosome arms and genes known to be frequently altered in tumorigenesis in primary and metastatic conjunctival melanomas (CoMs). METHODS DNA (200 ng) was analyzed by three multiplex ligation-dependent probe amplification assays (P027 uveal melanoma, P036 human telomere, and P206 spitzoid melanoma). RESULTS Eight of 16 primary tumor samples and 4 of 6 metastatic samples showed BRAF V600E gene mutations. CDKN1A and RUNX2 (both 6p21.2) were amplified in 11 and 16 of 21 primary CoMs, respectively. In metastatic CoMs, MLH1 (3p22.1) and TIMP2 (17q25.3) were frequently amplified, and MGMT (20q26.3) and ECHS1 (10q26.3) were frequently deleted. The BDH (3q), FLJ20265 (4p), OPRL1 (20q), and PAO (10q) genes, representing the telomeres of their respective chromosome arms in the P036 assay, were frequently amplified in metastatic CoMs. No statistically significant associations were identified between BRAF mutation or CDKN1A or RUNX2 amplification and sex, age, histologic cell type, or patient survival. CONCLUSIONS No copy number changes were associated exclusively with metastatic CoMs. However, further investigation of the role of CDKN1A and RUNX2 in CoMs development and that of MLH1, TIMP2, MGMT, and ECHS1 in metastatic CoMs is warranted. Validation of the observed gene and chromosome arm copy number changes in a larger cohort of primary and metastatic CoMs is necessary to identify the patients at highest risk for CoMs metastasis.

Artificial Neural Networks Estimating Survival Probability after Treatment of Choroidal Melanoma

PURPOSE To describe neural networks predicting survival from choroidal melanoma (i.e., any uveal melanoma involving choroid) and to demonstrate the value of entering age, sex, clinical stage, cytogenetic type, and histologic grade into the predictive model. DESIGN Nonrandomized case series. PARTICIPANTS Patients resident in mainland Britain treated by the first author for choroidal melanoma between 1984 and 2006. METHODS A conditional hazard estimating neural network (CHENN) was trained according to the Bayesian formalism with a training set of 1780 patients and evaluated with a test set of another 874 patients. Conditional hazard estimating neural network-generated survival curves were compared with those obtained with Kaplan-Meier analyses. A second model was created with information on chromosome 3 loss, using training and test sets of 211 and 140 patients, respectively. MAIN OUTCOME MEASURES Comparison of CHENN survival curves with Kaplan-Meier analyses. Representative results showing all-cause survival and inferred melanoma-specific mortality, according to age, sex, clinical stage, cytogenetic type, and histologic grade. RESULTS The predictive model plotted a survival curve with 95% credibility intervals for patients with melanoma according to relevant risk factors: age, sex, largest basal tumor diameter, ciliary body involvement, extraocular extension, tumor cell type, closed loops, mitotic rate, and chromosome 3 loss (i.e., monosomy 3). A survival curve for the age-matched general population of the same sex allowed estimation of the melanoma-related mortality. All-cause survival curves generated by the CHENN matched those produced with Kaplan-Meier analysis (Kolmogorov-Smirnov, P<0.05). In older patients, however, the estimated melanoma-related mortality was lower with the CHENN, which accounted for competing risks, unlike Kaplan-Meier analysis. Largest basal tumor diameter was most predictive of mortality in tumors showing histologic and cytogenetic features of high-grade malignancy. Ciliary body involvement and extraocular extension lost significance when cytogenetic and histologic data were included in the model. Patients with a monosomy 3 melanoma of a particular size were predicted to have shorter survival if their tumor showed epithelioid cells and closed loops. CONCLUSIONS Estimation of survival prognosis in patients with choroidal melanoma requires multivariate assessment of age, sex, clinical tumor stage, cytogenetic melanoma type, and histologic grade of malignancy.

Partial Logistic Artificial Neural Network for Competing Risks Regularized With Automatic Relevance Determination

Time-to-event analysis is important in a wide range of applications from clinical prognosis to risk modeling for credit scoring and insurance. In risk modeling, it is sometimes required to make a simultaneous assessment of the hazard arising from two or more mutually exclusive factors. This paper applies to an existing neural network model for competing risks (PLANNCR), a Bayesian regularization with the standard approximation of the evidence to implement automatic relevance determination (PLANNCR-ARD). The theoretical framework for the model is described and its application is illustrated with reference to local and distal recurrence of breast cancer, using the data set of Veronesi (1995).

Modelling survival after treatment of intraocular melanoma using artificial neural networks and Bayes theorem

This paper describes the development of an artificial intelligence (AI) system for survival prediction from intraocular melanoma. The system used artificial neural networks (ANNs) with five input parameters: coronal and sagittal tumour location, anterior tumour margin, largest basal tumour diameter and the cell type. After excluding records with missing data, 2331 patients were included in the study. These were split randomly into training and test sets. Date censorship was applied to the records to deal with patients who were lost to follow-up and patients who died from general causes. Bayes theorem was then applied to the ANN output to construct survival probability curves. A validation set with 34 patients unseen to both training and test sets was used to compare the AI system with Coxs regression (CR) and Kaplan-Meier (KM) analyses. Results showed large differences in the mean 5 year survival probability figures when the number of records with matching characteristics was small. However, as the number of matches increased to > 100 the system tended to agree with CR and KM. The validation set was also used to compare the system with a clinical expert in predicting time to metastatic death. The rms error was 3.7 years for the system and 4.3 years for the clinical expert for 15 years survival. For < 10 years survival, these figures were 2.7 and 4.2, respectively. We concluded that the AI system can match if not better the clinical experts prediction. There were significant differences with CR and KM analyses when the number of records was small, but it was not known which model is more accurate.

Immunohistochemical assessment of mitotic count in uveal melanoma

Background/Aims:  The mitotic count of uveal melanomas correlates with the risk of metastatic death, but with haematoxylin and eosin (H&E)‐stained sections, it can be difficult to identify mitotic figures (MF) reliably. We investigated whether this measurement could be enhanced by immunohistochemistry, using the mitosis‐specific marker Phospho‐Histone H3 Ser10 (PHH3).