1
428 Computers in cancer management The National Library of Medicine at Bethesda, Maryland, USA, was the venue for a futuristic look at Computer Applications for Early Detection and Staging of Cancer. Organised by the Early Detection Branch of the National Cancer Institute, the meet- ing was attended by a widely varied group of 70 people with medical and scientific backgrounds who came to- gether on July 29-30 to listen to and to discuss 26 presentations. The three themes were radiological image enhancements, laboratory data analysis, and cancer staging. Although controversial, the development of neural network computer analysis methods (of which there are several different types) seems to be producing the means by which complex constell- ations of information can be arranged and rearranged by a process of "net- work training" such that the resultant "conclusion" has value for clinical decision-making. The background was well described by W G Baxt (University of California, San Diego Medical Center), whose abstract in the conference programme provided a useful list of references. The use of artificial neural networks for advanced pattern recognition goes back over 25 years but progress was delayed by the discovery that single-layer neural net- works could not solve even the most basic non-linear problems. The multi- layer network has, however, been shown to be a potent aid to the analysis of complex problems. It has been applied widely to clinical medicine, especially over the past two years. It has been used to detect aromas, and to simulate somatosensory systems and colour vision. Networks have been used to analyse a wide range of imaging data (eg, thermal images, : magnetic resonance scans, radio- graphs, ultrasound images), and wave- form data (eg, electrocardiograms, electroencephograms, arterial pres- sure wave-forms), cytological find- ings, and pharmacokinetic properties of drugs. The ability of artificial neural networks to perform accurately even when provided with partly inaccurate raw data, makes them valuable for clinical diagnosis. They have been applied fairly successfully to the diag- nosis of skin lesions, appendicitis, v prediction of the onset of diabetes mellitus, back pain, and dementia, and the analysis of anterior chest pain. The basic objectives in using neural networks for analysis of radiological imaging and laboratory data are not only to enhance the quality of the information but also to study the relation within the data such that the sensitivity, specificity, and positive predictive value of the resultant con- clusion can be identified and be used as a basis for improved management. By contrast, the potential of neural net- works for predicting outcome of treatment for a particular patient or group of patients on the basis of prognostic factors has not yet been realised. Furthermore, the prognosis for patients with intermediate stage cancer as defined by the TNM system (ie, stages II and III) cannot be pre- dicted reliably. Hence the American Joint Committee on Cancer (AJCC) has recently committed itself to the study of new systems for deriving more accurate predictions of outcome. The discussion at this meeting con- cluded that neural networks are worthy of study in regard to the refinement of staging systems for pre- dicting outcome from analysis of multiple factors. However, many problems remain with the use of neu- ral networks or the more classic statis- tically based approach to the predic- tion of outcome (Cox regression and Bayes’ probabilistic theorem): the management of censored data, in- formation scaling, mixed data types, and missing and "messy" data are examples of the issues that need to be critically analysed for all predictive methods. A most challenging subject is the possibility that some prognostic factors have different effects on out- come depending on the time vector of the estimation. For example, a factor may have a strong influence on out- come for the first two years, and then its effect may disappear while others take over in prognostic weight. This inconsistent effect on prognosis is called "non-monotonicity" and it represents a methodological challenge for all prognosticsystems. Although this meeting was at times somewhat disjointed, it represented an important milestone in the advancement of medical information systems that we will all be using soon. The exploitation of neural networks is likely to play an important part in the : developments. L Peter Fielding Hitch in tobacco and alcohol advertising ban Moscow City Council last month passed a total ban on tobacco and alcohol advertis- ing. The snap decision led to an outcry, not only from Moscow’s new advertising agen- cies but also from legal experts, for Russia’s new laws on entrepreneurial activities allow businesses to engage in any activities other than those specifically prohibited by law. And Moscow, being only a "region" within the Russian Federation, cannot issue its own laws. The City Council is empowered to issue local ordinances, but these must not contravene federal legislation. : In a draft "Municipal Programme", Moscow City Council had envisaged bans on tobacco and alcohol advertisements in specific locations, restrictions on adver- tisement size, and a surcharge on advertis- ing fees for these products. Last month’s decision, however, envisaged a total ban not only on the products, but also on all other goods marketed under the same name publicity for foodstuffs produced by a tobacco company is good publicity for its cigarettes. : The advertising agencies hint that the new ruling was prompted not so much out of concern for public health but as a form of reprisal for not responding to the City’s request to the business community to raise 3 million roubles for holidays for children with alcoholic parents. A Moscow ban will reach far outside the City limits, since most of Russia’s national newspapers and the main TV channels originate in Moscow. Foreign newspapers, also, will be affected, but it is difficult to see how such a ban could be implemented. Russian federal law is, in fact, slowly moving towards a ban on tobacco and alcohol advertising in the media. Such a ban is envisaged in the new "Fundamentals of Legislation on Public Health", which recently passed their second reading in Parliament. But, as currently worded, the ban would apply to a product, not to a firm’s name. Penalties for offenders, it is said, will be embodied in a future amend- ment to the existing Mass Media Law, and in a projected Law on Advertising. Mean- while, in Moscow, cigarette posters are due to come down this month. Vera Rich

CONFERENCE: Computers in cancer management

Embed Size (px)

Citation preview

Page 1: CONFERENCE: Computers in cancer management

428

Computers in cancer management

The National Library of Medicine atBethesda, Maryland, USA, was thevenue for a futuristic look at ComputerApplications for Early Detection andStaging of Cancer. Organised by theEarly Detection Branch of the

National Cancer Institute, the meet-ing was attended by a widely variedgroup of 70 people with medical andscientific backgrounds who came to-gether on July 29-30 to listen to and todiscuss 26 presentations.The three themes were radiological

image enhancements, laboratory dataanalysis, and cancer staging. Althoughcontroversial, the development ofneural network computer analysismethods (of which there are severaldifferent types) seems to be producingthe means by which complex constell-ations of information can be arrangedand rearranged by a process of "net-work training" such that the resultant"conclusion" has value for clinical

decision-making. The backgroundwas well described by W G Baxt(University of California, San DiegoMedical Center), whose abstract in theconference programme provided auseful list of references. The use ofartificial neural networks for advancedpattern recognition goes back over 25years but progress was delayed by thediscovery that single-layer neural net-works could not solve even the mostbasic non-linear problems. The multi-layer network has, however, beenshown to be a potent aid to the analysisof complex problems. It has been

applied widely to clinical medicine,especially over the past two years. Ithas been used to detect aromas, and tosimulate somatosensory systems andcolour vision. Networks have beenused to analyse a wide range of

imaging data (eg, thermal images, :magnetic resonance scans, radio-

graphs, ultrasound images), and wave-form data (eg, electrocardiograms,electroencephograms, arterial pres-sure wave-forms), cytological find-

ings, and pharmacokinetic propertiesof drugs. The ability of artificial neuralnetworks to perform accurately evenwhen provided with partly inaccurateraw data, makes them valuable forclinical diagnosis. They have beenapplied fairly successfully to the diag-nosis of skin lesions, appendicitis, v

prediction of the onset of diabetes

mellitus, back pain, and dementia, andthe analysis of anterior chest pain.

The basic objectives in using neuralnetworks for analysis of radiologicalimaging and laboratory data are notonly to enhance the quality of theinformation but also to study therelation within the data such that the

sensitivity, specificity, and positivepredictive value of the resultant con-clusion can be identified and be used asa basis for improved management. Bycontrast, the potential of neural net-works for predicting outcome oftreatment for a particular patient orgroup of patients on the basis of

prognostic factors has not yet beenrealised. Furthermore, the prognosisfor patients with intermediate stagecancer as defined by the TNM system(ie, stages II and III) cannot be pre-dicted reliably. Hence the AmericanJoint Committee on Cancer (AJCC)has recently committed itself to thestudy of new systems for derivingmore accurate predictions of outcome.The discussion at this meeting con-cluded that neural networks are

worthy of study in regard to therefinement of staging systems for pre-dicting outcome from analysis of

multiple factors. However, many

problems remain with the use of neu-ral networks or the more classic statis-

tically based approach to the predic-tion of outcome (Cox regression andBayes’ probabilistic theorem): the

management of censored data, in-

formation scaling, mixed data types,and missing and "messy" data are

examples of the issues that need to becritically analysed for all predictivemethods. A most challenging subjectis the possibility that some prognosticfactors have different effects on out-come depending on the time vector ofthe estimation. For example, a factormay have a strong influence on out-come for the first two years, and thenits effect may disappear while otherstake over in prognostic weight. Thisinconsistent effect on prognosis iscalled "non-monotonicity" and it

represents a methodological challengefor all prognosticsystems.

Although this meeting was at timessomewhat disjointed, it represented animportant milestone in theadvancement of medical information

systems that we will all be using soon.The exploitation of neural networks islikely to play an important part in the

: developments.

L Peter Fielding

Hitch in tobacco andalcohol advertising banMoscow City Council last month passed atotal ban on tobacco and alcohol advertis-

ing. The snap decision led to an outcry, notonly from Moscow’s new advertising agen-cies but also from legal experts, for Russia’snew laws on entrepreneurial activities allowbusinesses to engage in any activities otherthan those specifically prohibited by law.And Moscow, being only a "region" withinthe Russian Federation, cannot issue itsown laws. The City Council is empoweredto issue local ordinances, but these must notcontravene federal legislation. :

In a draft "Municipal Programme",Moscow City Council had envisaged banson tobacco and alcohol advertisements inspecific locations, restrictions on adver-tisement size, and a surcharge on advertis-ing fees for these products. Last month’sdecision, however, envisaged a total ban notonly on the products, but also on all othergoods marketed under the same namepublicity for foodstuffs produced by atobacco company is good publicity for itscigarettes. :

The advertising agencies hint that thenew ruling was prompted not so much outof concern for public health but as a form ofreprisal for not responding to the City’srequest to the business community to raise3 million roubles for holidays for childrenwith alcoholic parents.A Moscow ban will reach far outside the

City limits, since most of Russia’s nationalnewspapers and the main TV channels

originate in Moscow. Foreign newspapers,also, will be affected, but it is difficult to seehow such a ban could be implemented.

Russian federal law is, in fact, slowlymoving towards a ban on tobacco andalcohol advertising in the media. Such aban is envisaged in the new "Fundamentalsof Legislation on Public Health", whichrecently passed their second reading in

Parliament. But, as currently worded, theban would apply to a product, not to afirm’s name. Penalties for offenders, it issaid, will be embodied in a future amend-ment to the existing Mass Media Law, andin a projected Law on Advertising. Mean-while, in Moscow, cigarette posters are dueto come down this month.

Vera Rich