Applications of an inexpensive microcomputer for the general radiologist

Applications of an inexpensive microcomputer for the general radiologist

Cumpurrrixd Rudiol. Vol. 7, No. Printed in the U.S.A. I, pp. 49-60. APPLICATIONS FOR 0730-4862/83/010049-12$03.00/O Pergamon Press Ltd 1983 OF AN...

1MB Sizes 0 Downloads 59 Views

Cumpurrrixd Rudiol. Vol. 7, No. Printed in the U.S.A.

I, pp. 49-60.

APPLICATIONS FOR

0730-4862/83/010049-12$03.00/O Pergamon Press Ltd

1983

OF AN INEXPENSIVE MICROCOMPUTER THE GENERAL RADIOLOGIST MICHAEL

Department

of Radiology/SGHR, (Rrceioed

9 March

L.

USAF

RICHARDSON Hospital.

1982; recriued,for

Mather

publication

AFB. CA 95655, U.S.A 16 June 1982)

Abstract-Computer programs of interest to the general radiologist have been written by the author and others for an inexpensive microcomputer. Most of these programs are in the public domain and include: computer-assisted diagnosis; computer-assisted medical education; Bayesian analysis, general statistical analysis; word processing; and radiographic image analysis, Previously, computerized functions such as these have been limited to those with access to a large institutional computer. Due to the general availability of inexpensive microcomputers, the average radiologist or group can now afford the many benefits a computer can add to the practice of radiology. Microcomputer Computer Computer assisted gence graphic report generation

diagnosis instruction Information

Bayesian analysis Decision Radiographic image analysis storage and retrieval

analysis Artificial Word-processing.

intelliradio-

INTRODUCTION

Computers have transformed radiology so vastly in the past decade that the term “radiology” itself may soon become obsolete and supplanted by the more inclusive term, “diagnostic imaging”. Diagnostic imaging is currently the most important area of radiologic computing and computed tomography (CT) is the epitome of this. However, the purpose of this paper is to introduce several useful, but less well known areas of medical computing, and to show how many of them can be implemented on an inexpensive microcomputer. A brief account of the history of the microcomputer will be offered, and several important terms will be defined. Examples from the medical literature and from the author’s personal experience will then be given of how the general radiologist can benefit from a personal computer. A SHORT

HISTORY

OF

COMPUTING

In 1944, there were no digital computers. The closest thing to one anywhere in the world was a huge digital calculator at Harvard University, nicknamed Mark 1, which used mechanical switching relays to make its computations. Two years later, in 1946, ENIAC (Electronic Numerical Integrator and Calculator), the world’s first electronic digital computer, was built at the University of Pennsylvania. Although it still contained 1500 relays, the bulk of the computations were performed by some 18000 vacuum tubes. It filled a large room, and took around 3 years to design and build [l]. Just as the forties saw the birth of the computer, each succeeding decade witnessed a quantum leap in computer technology. By 1950, the computer had emerged as a general tool of government, science, industry, education, and the military; and the vacuum tubes had been replaced by smaller and more efficient transistors. The mid-sixties saw the introduction of the integrated circuit, which compressed the equivalent of thousands of transistors, resistors, diodes, and capacitors onto a silicon chip one-eighth inch square. In the seventies, large-scale integration (LSI) was introduced, and an entire computer could be compressed onto a single chip. The first personal microcomputer used such a chip, and was first sold in kit form in 1975. by a now defunct company known as Mits. Over one dozen other companies were producing microcomputers by 1976. By 1980, one could choose from among several dozen manufacturers, and buy the computer fully assembled and tested. Hundreds of accessories are currently available, including speech recognition units, powerful graphics analysis units. business programs, and music synthesizers. 49

50

MICHAEL L. RICHAKDSON

The cost of a single integrated circuit chip has not changed significantly since the mid-sixties’ price of 5 to 50 dollars. The capabilities of a single chip, however, have increased by a factor of 10000 and experts do not foresee any immediate end to this trend. They predict that by the end of the century, one chip, several millimeters on a side, will have approximately 10000000 internal memory cells, and be able to perform around 20000000 operations per second. Such a chip could almost single-handedly perform all of the computational and storage functions of a present-day CT unit. GLOSSARY The field of computing is just as replete with jargon as is any field of medicine. In medicine, however, many terms are familiar to the general public. Therefore, many laymen, especially aficionados of medical soap operas, can follow the gist of a conversation between physicians. “Computerese”, however, is several times removed from the common human experience, and can be extremely daunting to most persons outside the field. However, radiologists have a great advantage over most other physicians, since the advent of CT has forced many of us to come to at least speaking terms with some of the esoterica of computing. This article will avoid all but the most basic details of computing science, and a glossary of several key terms will now be given. Computer. A device consisting of memory and logic units, which can carry out a list of instructions. Microprocessor. An integrated circuit chip which contains a complete computer within it. However, to be useful, this chip must be combined with other items such as a power supply, extra memory, and input/output devices such as a television screen, keyboard, or printer. Microcomputer. A combination of a microprocessor, extra memory, input/output devices, any desired accessories, and a box to put them in. Memory. A device which can store a piece of information for later recall. This information may be a number, an English phrase, a computer program, or an index to a radiology teaching file. Memory comes in two forms, volatile and non-volatile. Volatile memory. Memory in which the stored information is lost forever once the power supply is turned off. This stored information can be changed at any time by the user. An integrated circuit chip which does this is called a RAM (Random Access Memory). Many pocket calculators contain one or more small memories such as this. Non-volatile memory. A type of memory in which the information will be retained when the power is turned off. An integrated circuit which does this is called a ROM (Read Only Memory). In general, this type of memory cannot be changed by the user. Pocket calculators also contain memory of this type to store simple constants, such as the value of rc. Magnetic memory. Magnetic tape is another type of non-volatile memory, and can take the form of reel-to-reel tape, cassette tape, or magnetic disks. In this type of memory unit, the information can be changed at will. Disk memory. Information is stored on thin magnetic disks by a special type of recorder, known as a disk drive unit. One can store or recall information much faster on a disk unit than a reel-to-reel tape unit. This is because its playback/record head is free to jump from track to track just like the stylus of a phonograph turntable. It can therefore completely avoid large intervening areas of irrelevant memory in its search for a desired datum. Disk memory comes in two basic forms, flexible disks and hard disks [2]. Flexible Disks. Also known as “floppy” disks, these are the cheapest form of disk storage. Many nuclear medicine computers and some CT units use this form of memory. The main disadvantages of this type of disk storage are limited storage capacity and vulnerability to dirt or moisture. Hard disks. These units employ a set of rigid magnetic disks which may be enclosed at the factory in a hermetically sealed container. This protects the disks and the playback/record head from the surrounding environment. Because of this greater protection, the playback/record head can be positioned much closer to the disk surface, which in turn allows much higher information densities to be recorded. Most CT units employ such a unit for storing the day’s runs. Byte. A basic memory unit in a computer. The familiar prefixes, kilo-, mega- and giga- are added for convenience when speaking of large blocks of memory. Most microcomputers have nearly instantaneous access to approximately 64 kilobytes of RAM or ROM memory inside their cabinets.

Applications

of a microcomputer

for the general

radiologist

51

(The actual number is 65 536 bytes, but when speaking of memory, the common practice is to round off to the nearest multiple of 8 kilobytes.) In more familiar terms, 64 kilobytes is sufficient to store about 12000 words of plain English text. It is also enough to hold a program sufficient to run a small CT unit. The flexible disks used by many microcomputers are about 5 inches in diameter, and can hold approximately 120 kilobytes of storage space on each side. This is roughly the equivalent of fifty type-written pages of text on each side. Several hard disk units are available for microcomputers which will store ten megabytes in a unit the size of a cigar box. This is equal to over four thousand pages of type-written text. Inexpmsiz~e. In the context of this article, an inexpensive microcomputer is one which costs between 1000 and 3000 US dollars. The price can vary widely, depending upon the manufacturer, the specific model, the amount of memory, and the number and type of accessories desired. Hurdwarr. The actual integrated circuit logic and memory chips in a computer. In general, the user cannot easily change or modify the hardware elements. S@wrr. The list of instructions, or program, which the user gives to the computer. These programs may be written by the user, the computer manufacturer, or a third-party software manufacturer.

REASONS

TO

BUY

A COMPUTER

There are many good reasons for a radiologist to buy a computer: (1) by handling tedious or routine tasks, it can free the radiologist for more difficult and challenging tasks; (2) by creative programming, even the interesting cases can be made more interesting; (3) appropriate programming can help increase the radiologist’s income; (4) computers can improve the patient’s care, and make the radiologist more useful to clinicians; (5) a computer can be invaluable in research; (6) computers can be useful to a radiologist away from work. These reasons will now be explored in greater detail, and multiple examples will be given of each one. These computer applications will be arbitrarily divided into seven categories: computer diagnosis; statistical analysis; computer assisted education; computer modeling and simulation; image analysis; business programming; and miscellaneous applications.

COMPUTER

DIAGNOSIS

AND

THERAPY

PLANNING

Computers can not only create the images we look at, as in CT, but they can also interpret the images and form diagnoses. This can be done in one of two ways. In the first method, the computer itself scans the image, decides which roentgen patterns are present, and then computes a diagnosis. This method has been used by Kruger et al. for the computer diagnosis of pneumoconiosis [3]. The second method requires a human film reader to first enter the roentgen findings into the computer, which then forms a diagnosis. This is the most common method of computer diagnosis described in the medical literature. A comprehensive bibliography of 827 such articles has been assembled by Wagner et al. [4]. The same decision making algorithms used to make the diagnosis can also be used by a computer to decide on an appropriate treatment plan. Houdek has described the use of a microcomputer in radiotherapy treatment planning [S]. Another computer program has been written for infectious disease consultation. A comparison between its performance and that of a panel of human experts has been reported by Yu et al. [6]. More and more radiologists are beginning to do therapeutic procedures, such as percutaneous transluminal angioplasty and catheter drainage techniques. Undoubtedly, computers will prove useful in the planning of these procedures as well. Radiographic diagnosis by computers will probably meet a mixed response from human radiologists. The prospect of having one’s professional income diminished by competition with a computer is unlikely to be an endearing one. At the same time, the possibility of using a computer to improve one’s diagnostic accuracy or to increase one’s efficiency is extremely attractive. In difficult cases, a computer’s total recall linked to our own more fallible memories could help insure that no significant possibilities are overlooked in forming a diagnosis. Physicians have a long tradition of embracing any new technology which improves patient care. Therefore, if computers can someday

MICHAEL L. RICHARDSON

52

do a reliable job of medical diagnosis, physicians will probably use them. The crucial question then becomes: how reliable is computer diagnosis? The reliability of a computer’s diagnosis is most dependent upon the accuracy of the diagnostic algorithm which it has been taught, and not upon the computer’s inherent computational abilities. Therefore, to understand diagnostic accuracy, one must understand something of the major algorithms currently in use. These methods, plus their pros and cons, have been discussed in detail both by Henschke et al. [7] and by Shortliffe et al. [S]. Therefore, only short summaries of these methods will be given in this paper, followed by examples of their use. STATISTICAL

PATTERN

RECOGNITION

Statistical pattern recognition techniques are currently in routine use in medicine. These techniques are used in many hematology laboratories by automatic cell counters. These machines are able to distinguish different types of white blood cells, and thus give automated differential counts. Such techniques can also be used to recognize the pattern of a particular disease in a patient’s clinical and radiologic findings. The linear discriminant function is a simple example of how this can be done. Each of the patient’s various findings are expressed in equation (1) by S(K), where K is the number of total findings. A weighting factor, a(K), is added for each finding, depending upon its relative importance to the other findings. All of the findings present in a given patient are then multipled by their weighting factors, and summed to give a new variable, Z. It is assumed that each different disease will have a different Z number. After the patient’s Z number has been computed, the diagnosis is made by finding the disease whose Z number comes the closest to that of the patient. z = a(l)S(l) + a(2)S(2) + .... + a(K)S(K).

(1)

This method was used by Freemon [9] in programming a computer to diagnose the cause of headaches. Though admittedly simple, his program correctly diagnosed 16 of 20 cases of headache. This program is easily adapted for use on a microcomputer. BAYESIAN

ANALYSIS

Another popular approach to computer diagnosis has involved the use of Bayes’ Theorem, which is shown in its general form in equation (2). This equation is used to calculate P(DiIS) which is the probability that disease i is present, given that the finding S is present. P(Di) is the incidence of disease i in our patient population, and P(SlDi) is the probability that finding S is present, given that the patient has disease i. The variable m represents the total number of diseases in the differential. p(D,lS)

=

mp(Di)p(sIDi)

C

j=l

(2)

p(Dj)p(slDj)

Although this equation is somewhat formidable in appearance, its practical application can easily be understood by a simple example. The interested reader is referred to excellent articles by Warner et al. [lo] and Patrick [l l] for a more detailed discussion. Let us assume that in working up a patient, we have narrowed the differential diagnosis down to one of two mythical diseases, which both occur in the general population with equal incidence. Upon consulting the literature about these two diseases, we learn some interesting facts about their roentgen findings, which are displayed in Table 1. From these facts, we can predict that, given a patient with Tibetan Yak fever, there is a 30% chance of finding parenchymal calcifications and a 65% chance of finding hilar adenopathy on the chest radiograph. Similar percentages are given for the other imaginary disease, prodsponder plague. These probabilities are nice to know, and are sometimes called prior, or a priori probabilities. However, it would be much more useful to have percentages which would predict which disease the patient has, given only that the chest radiograph shows a particular finding. This type of probability is called posterior or a posteriori probability.

Applications

Table

Disease Tibetan fever

of a microcomputer

1. Probability

Incidence of disease (%)

for the general

of a finding,

53

radiologist

given these diseases

Incidence of pulmonary parenchymal calcifications (%)

Incidence of Hilar adenopathy (%)

Yak

Prodsponder plague

50

30

65

50

50

20

Unfortunately, while the radiologic literature abounds with listings of prior probabilities, it is uncommon to find articles quoting the more useful posterior probabilities. Bayes’ Theorem helps us out of this dilemma by giving us a method for calculating a posterior probability once we know a prior probability. Without putting the reader through the actual computations, let us look at the results we get by applying Bayes’ Theorem to the probabilities in Table 1. These results are shown in Table 2. If our patient presents with adenopathy only, he will have a 76% chance of having Tibetan yak fever, and a 24% chance of having prodsponder plague. On the other hand, a patient having neither adenopathy nor calcifications will have a 38% likelihood of having yak fever and a 62% chance of having plague. This example only considers two possible diseases and two possible roentgen findings. However, equation (2) will allow one to write a diagnostic program which can consider as many separate symptoms, roentgen findings, and diseases as desired. This method has been used for the computer analysis of bone tumors [12], renal masses [13], gastric ulcers [14], solitary pulmonary nodules [15], congenital heart disease [16], and pleuritic chest pain [17]. All of these programs can easily be implemented on a microcomputer. DECISION

THEORY

A third common technique used in computer diagnosis is the decision tree, an old friend to many radiologists. Decision trees represent a very compact way of displaying the flow of logic in a diagnostic workup, even when many diagnostic modes and diseases are under consideration [18]. It is not necessary to have a computer in order to use a decision tree. However, if the tree has many branches, and the choice of paths involves the calculation of probabilities, a computer can be a great timesaver. A computer program built around such an algorithm simply asks for new data at each branch point in the algorithm, and then chooses the appropriate branch in the tree. Bleich [19] has used this method to write a sophisticated program for the diagnosis of acid-base disorders. This technique also lends itself well to use on a microcomputer. ARTIFICIAL

INTELLIGENCE

A fourth approach used in computer diagnosis involves the use of artificial intelligence (AI) techniques. AI is the portion of computer science concerned with making computers produce intelligent action. While much of computer science is involved with manipulation of numbers and

Table 2. Probability

Disease Tibetan fever

Parenchymal calcifications only (%)

of a disease, given these findings Hilar adenopathy only (%)

Both findings present (%)

Neither finding present (%)

Yak

Prodsponder plague

38

76

66

38

62

24

34

62

54

MICHAEL L. RICHARDSON

equations, AI is more concerned with manipulation of symbols and ideas. Many of the problems which AI attempts to solve are not easily reducible to a precise mathematical formula. Medical diagnosis is an excellent example of such a problem. Although such problems can occasionally be solved by brute force calculations, the best solutions are usually arrived at by a combination of subtlety and common sense. Much of the current research in AI is therefore an attempt to understand how a human expert analyzes and solves a difficult problem. This inexact science is known as heuristics, or “the art of good guessing” [20]. One common AI technique involves the use of production rules. A production rule is a logical statement similar to the syllogism of Aristotelian logic. An example of a production rule which might be used for the diagnosis of renal masses is shown below. If: (1) the mass has a large fatty component, and (2) macroaneurysms are demonstrated on angiography, and (3) the mass is solid Then: there is strong evidence that the mass is an angiomyolipoma. Many such production rules can be constructed, and entered into the computer. The computer then takes the available clinical and radiological findings, sifts through the pertinent production rules, and makes an appropriate diagnosis. A program using this method has been described by Yu et al. [6] and is known as MYCIN. MYCIN was developed to emulate an infectious disease expert in the diagnosis and treatment of meningitis. Other AI techniques have been used by Pople and Myers [21] to construct a program known as INTERNIST. INTERNIST is designed to perform the awesome task of diagnosis of all diseases in internal medicine. It has been remarkably successful to date, and is able to correctly diagnose a large percentage of complex cases selected from clinical pathological conferences in the major medical journals. In contrast to the other techniques of computer diagnosis described in this paper, programs such as MYCIN and INTERNIST are not easily adaptable for use on a microcomputer. This is due to their extremely large size and complexity. These programs use sophisticated AI techniques which allow the user to converse with the program in plain English. This alone uses a great deal of computer memory. These programs are so complex that they run slowly even on large institutional computers. Despite all this, one can still use these programs with the aid of a microcomputer. To do this, the microcomputer is linked to the larger computer via the telephone lines and acts as a remote terminal for the larger one. This paper will further explore the idea of computer-to-computer linkage by telephone when business uses of microcomputers are discussed. In summary, many techniques have been used in computer diagnosis, often with astonishing success. In fact, the degree of success achieved thus far suggests that computer diagnosis can become as accurate and reliable in certain fields as a human expert within the next 20 years. If this comes to pass, we must have answers to some tough questions. For example, if such computer programs come into widespread use, who will accept the blame for a misdiagnosis-the physician or the computer programmer? If the physician disagrees with the computer and treats a patient in a different manner, will he be liable for mal-practice if he was wrong? Will the Federal Drug Administration be responsible for approving new diagnostic computer programs for general use? Such questions are not easily answered, and will probably confront us in the next two decades. However, even if computers never attain the level of medical expertise currently enjoyed by human physicians, the attempt will have been worthwhile. Dr Edward Feigenbaum, of the Department of Computer Science at Stanford University, has said that the fact that computers can do diagnosis is only of secondary importance. The most important point is that the unwritten rules of medical knowledge by which experts operate will finally be clarified and amassed in one place. COMPUTER-ASSISTED

INSTRUCTION

The ideal classroom was once defined as one teacher and one student sitting on opposite ends of a log. In educating our radiology residents and technicians, we have often drifted quite far from this ideal. We have done this for many reasons, including lack of staff, lack of funds, and lack of time for teaching. The problem is compounded by other facts. Not all of our teachers are good teachers. Students learn at different rates, and have different lacunae to fill in their knowledge.

Applications of a microcomputer

for the general radiologist

55

Educators have long hoped that computers would revolutionize education. In fact, Frenzel [22] traces the concept of computer-assisted instruction (CAI) back to 1924, when Dr Sidney Pressey invented a machine which would grade multiple choice tests. Teaching machines were a direct descendant of this idea, and were developed in the late 1950’s and early 1960’s. These machines would present a single concept to students, and then immediately test them on it. This small tid-bit of text and question is called a frame, and numerous frames are linked together in a sequential manner to form the lesson. This form of programmed teaching has been presented in book form [23] as well as on teaching machines, and in the early 1960’s was first implemented on computers. These early experiments demonstrated that CA1 was indeed an effective way to teach students. However, it was not shown to be any more effective than traditional methods of teaching. Moreover, the size and expense of computers in the 1960’s made them an impractical way of presenting the material to the student. The development of first minicomputers, and then microcomputers, significantly lowered the cost of CAI. However, Frenzel feels that it is still “inefficient to present material on a US $5OCLlOOOmachine when a low-cost book can be used to present the same material”. Another objection to CA1 has been that few quality education programs have been available for the general public. Although individual companies and universities have compiled their own libraries of such programs, it has not been easy to acquire copies for personal use. Recently. however, numerous small software companies have been producing programs for microcomputers, and this objection will soon be obsolete. A further objection to CA1 is that the frame method of teaching is rigid, and allows only limited feedback between the student and the program. Indeed, many frame-oriented CA1 programs do tend to treat all students alike, regardless of how much of the material they already know. Current research in this field is attempting to solve these problems by means of the artificial intelligence (AI) techniques mentioned earlier in this article. One such program, called GUIDON, is being developed at Stanford University [24]. This program is able to converse with the student in plain English, and can tailor a teaching session to the abilities and previous knowledge of the student. If the student does not understand how a certain answer was derived, he can query the computer in English. This program will then walk him through the logical steps it used to arrive at the answer. Conventional CA1 programs can provide this feature to some extent, but only to the degree that the programmer has anticipated any questions that the student might ask. GUIDON’s advantage is that it “understands” what the student is asking, and it attempts to intelligently construct an answer to that question. After devoting the last few paragraphs to what is wrong with CAL it is time to discuss what is right with it. It is appropriate to first address the issue of the practicality of CAI. Buying a microcomputer for CA1 alone, indeed, may not be a cost-effective way to teach or to learn. However, if one buys the microcomputer primarily for other reasons, then the ability to use it for CA1 becomes “icing on the cake”. Another important point is that even though CA1 may be no more effective than reading a book, it is often a great deal more interesting. Since student motivation is often an impediment to learning, anything which makes studying more interesting and exciting is certainly advantageous. One has only to consider the addictive potential of current computer games to realize the popularity that well-designed CA1 could have. Critics of CA1 frequently overlook its ability to teach via simulation. Once an appropriate mathematical model of a process has been constructed, a computer can put the student through extremely life-like exercises. An excellent example of this is in the field of aviation. The United States Air Force and many civilian airlines regularly put their pilots through hours of instruction on a flight simulator. This process is also employed by the National Aeronautics and Space Administration for training its astronauts. A chief benefit of teaching pilots by simulation is that the student can make errors without destroying an expensive airplane. The medical analogue of the expensive airplane is the patient, whose safety is beyond price. It would seem logical to train novice physicians using these same techniques. Radiologists frequently perform potentially lethal procedures, such as cerebral angiography or percutaneous needle biopsy of pulmonary lesions. By means of creative use of computer graphics and simulation, it should be possible to teach residents good catheter and biopsy technique before letting them practice with real patients. For a good introduction to the field of physiological simulation with a microcomputer, the reader is directed to Randall’s excellent book c251.

56

MICHAEL L. RICHARDSON

CA1 is currently an active area of medical research, and the National Library of Medicine has compiled a bibliography of recent literature on this topic [26]. Programs have been developed for the teaching of roentgen anatomy [27], the diagnosis and treatment of respiratory [28] and rheumatic diseases [29], embryology [30], differential diagnosis [31], thyroid disease [32], and many other subjects. At least one commercial vendor (Milliken Communications Corporation, St Louis MO) is offering Category I Continuing Medical Education materials in the form of a microcomputer program. Although they do not currently offer these materials for radiology, they are available for surgery, internal medicine, psychiatry, and urology. Numerous programs are commercially available which allow one to construct one’s own CA1 lessons. Such a program has been written in the BASIC language by the author of this article, and is in the public domain. A listing of this program will be provided to anyone sending a self-addressed, stamped envelope. STATISTICAL

ANALYSIS

Many radiologists view the study of statistics as drudgery, and soon relegate it to oblivion, once they have passed their written boards, However, the everyday practice of radiology requires the use of some basic statistical tools. A good example of this is in the determination of a child’s bone age. In their reports, many radiologists not only give an estimated bone age, but also include an estimate of how much the bone age differs from the chronological age. This difference is usually expressed in terms of standard deviations from the mean, and can be determined by referring to a standard reference table. This can be rather tedious work, especially when one has a whole stack of such films to interpret. This task can be made much less onerous by delegating all of the calculations to a simple computer program. The author has written such a program around the Elgenmark method of determining bone age [33]. To use this program, the radiologist counts the number of ossification centers in the hand, wrist, elbow, shoulder, hip, knee, ankle and foot. These are entered into the computer which automatically sums them. It then calculates the expected bone age, and prints the result, plus the number of standard deviations by which it differs from the chronological age. If other data are given, such as the parents’ heights and the child’s current weight and height, the child’s adult height can be predicted using the method of Roche et al. [34]. The radiologist interested in research can find many further uses for computerized statistical analysis. Standard deviations, linear regression coefficients, chi-square tests, and the Student’s t-test are some of the functions which can easily be implemented on a microcomputer. Although such programs are not hard to write, many of the more useful ones have been compiled into book form [35]. The use of such “canned” programs is a great time saver, especially for the neophyte programmer. A computer can also be a great help for anyone trying to document the efficacy of a radiographic test. As recently demonstrated by Bell [36], efficacy is an elusive quantity to define. Once it is defined, it can be very difficult to quantify. In order to help do this, one may have to invoke techniques such as decision theory [37], receiver operating characteristic (ROC) analysis [38,39], Bayesian analysis [40], and discriminant analysis [41]. All of these techniques can be implemented on a microcomputer without difficulty. IMAGE

ANALYSIS

Once a radiographic image has been digitized, wondrous things can be wrought with it by a computer. By varying the window settings on CT, we can see unparalleled soft tissue detail. By summing a series of gated gamma camera images, we can non-invasively view the left ventricle, and measure its ejection fraction. With the aid of digital-subtraction techniques and fluoroscopy, we can see the aortic arch and great vessels after an i.v. bolus of contrast. Although most of these functions are usually performed by a minicomputer, there is no intrinsic reason why they cannot be done by a simple microcomputer. Unfortunately, the video display which comes with most microcomputers lacks sufficient bandwidth to give the level of resolution seen on commercial instruments [42]. It is possible to link a high-grade video monitor to a microcomputer, but such monitors usually cost several times as much as most microcomputers. Microcomputers also tend to be much slower than

Applications

of a microcomputer

for the general

radiologist

57

the minicomputers used in commercial instruments. However, if speed is not a prime concern, and laboratory-grade resolution is not necessary, a microcomputer can perform several useful tasks for the radiologist. The radiologist is frequently asked to describe the size of a lesion. While third and fourth generation CT machines now have sophisticated software for doing this, older CT units lack such options. By means of a microcomputer, a radiologist can now measure distances and areas not only on CT images, but also on ultrasound images and even plain films. To achieve this, one must add an accessory known as a graphics tablet to the microcomputer. Depending on the quality desired, these can be acquired for US $20&700. A typical unit consists of a calibrated tablet with an electronic pen attached to it. In use, a film is placed on the tablet, and the pen is placed over the desired spots on the film. The computer then calculates the distance between them. By tracing around a structure, such as a left ventriculogram, the computer will automatically calculate the area of the ventricle on that projection. By tracing around the uterus on a midline ultrasound scan, the total uterine volume can be automatically calculated. Another interesting application of inexpensive computer graphics is in the area of three-dimensional (3D) display of CT and ultrasound images. High resolution 3D displays of the cervical and lumbar spine have been created with a minicomputer by Herman et al. [43]. However, if extremely high resolution is not necessary, an inexpensive microcomputer is entirely adequate. The author is currently perfecting a program which can display the outlines of CT images in 3D. To achieve this, one first traces around the desired structures on several sequential scans. The microcomputer then displays them in a 3D projection on the screen. By manipulating simple game paddle controls, one can rotate the object in several dimensions, and zoom toward or away from the object. It is felt by the author that such a program would be most useful in the teaching of anatomy. For example, by looking at two-dimensional images alone, radiology residents often have difficulty appreciating all of the 3D changes wrought by a cerebral mass lesion. With the aid of a 3D display, these changes can be grasped more easily. Such displays could also be useful in appreciating the anatomy of difficult areas in the abdomen, such as the lesser sac. BUSINESS

USES

OF

A MICROCOMPUTER

Most of us do not go through our radiology training because of a fascination with cost accounting or filling out insurance forms. Unfortunately, these necessary evils of our profession seem to take up a disproportionate share of our time. Fortunately, we can apply the same solution to this problem that big businesses use for their paperwork problems. While most large businesses handle their payrolls, billing, and other bookkeeping functions by means of a large computer, the average radiology practice can make do quite nicely with a microcomputer. A radiology practice with a very large workload may need the extra speed and memory capacity that a more expensive minicomputer can provide. However, buying such a system requires a rather large investment of not only money, but of time as well. It may take weeks before one’s staff is familiar with the equipment, and all of the “bugs” are worked out of the system. The decision of whether or not to buy a computer is comparatively easy, compared to the decision of which brand of computer to buy. The initial difficulty is that many radiologists know very little about computers. While most of us can feign a semblance of sophistication when we face a car dealer, or a film detail man, we are quickly intimidated by the computer salesman. It is even more confusing when competing computer salesmen make contradictory claims about their equipment and that of their competitors. The problem is further aggravated by the astonishing growth in the computer industry. When today’s state-of-the-art equipment may be obsolete in three years, it is hard to know what to buy. Once all of the above problems have been solved, one must then face the possibility of computer malfunction. Unfortunately, computer systems can be just as prone to equipment failure and excessive down time as any conventional radiographic equipment. A microcomputer can be the solution to these many problems. For an initial outlay of US $2000-3000, one can build a simple business system around a microcomputer and “get one’s feet wet”. By hands-on experience, one quickly learns “computerese”, as well as the shortcomings and advantages of a particular computer system. In this manner, it will quickly become obvious if a microcomputer is adequate to handle one’s bookkeeping needs. If it is not, one can quickly upgrade

58

MICHAEL L. RICHARDSON

to a larger computer, and use the smaller one for some other purpose, such as word-processing. This strategy has another important side benefit. By learning to run a microcomputer, one can learn enough to be able to shop intelligently for a larger system. Let us now discuss some of the business applications of a microcomputer. The aspect of accounting, billing, and payroll will not be mentioned further, other than to say that a great deal of business software for microcomputers is already on the market, and ready for use. Instead, less obvious uses for a computer will be discussed, such as film tracking, report generation, data storage, and patient scheduling. No radiology department is free from the curse of missing films, and even computers will probably not completely eradicate this problem. However, a properly programmed computer can take care of a great deal of the tedious paperwork involved in film tracking. When a film is checked out, the necessary data can be entered into the computer by some relatively painless means. One such way is by using the same bar code used in supermarkets. This can be placed on all film jackets, and can be read with a special wand. Once the computer knows where the films are, it can periodically generate a list of overdue items. Appropriate actions can then be taken against those who sequester films. If the computer has a very large external memory unit, one may consider having the computer keep track of every single study performed on every single patient. With an even larger memory, one could store all of the dictated reports as well. Such computerized film tracking systems are currently commercially available on minicomputer systems. While I know of no such software which is specifically written for a microcomputer, there are many data-base management programs written for microcomputers which could be adapted for this use. Word processing is another useful task for a microcomputer. With one of the many programs currently available, a microcomputer can be transformed into a “smart typewriter”. With such a tool, one can compose letters, X-ray reports, and scientific papers on the screen of the computer terminal. This paper was written on a microcomputer with just such a program. Most of these programs have powerful editing functions built into them so that one can correct errors, delete entire sentences and paragraphs, move blocks of text around, and insert new words into a line. Once the text is to one’s liking, it can be stored on a magnetic disk or printed. Many powerful features are usually built into the printing subroutines of a word processor. Such features allow one to choose the placement of margins, the spacing between lines, the number of spaces to indent for new paragraphs, and the number of lines one would like printed on each page. Another useful feature is the ability to perform right and left hand justification of the text. This means that the printed text will line up evenly at both the right and left margins. Once an item has been stored on disk, it can be recalled at any time for revisions or for extra copies. One more use for a computer is in the generation of X-ray reports. While several computerized X-ray report generators are now on the market, they are beyond the needs and the means of the small to mid-sized radiology office. A microcomputer could be used to handle much of this work. The necessary software could be written either by the user or by a professional programmer. Even if such a system were only used for typing “normal” reports, it could make one’s stenographer much more efficient. Microcomputers can also be used to help produce optimum patient flow through one’s department. One approach involves scheduling the patients to fit the equipment [44]. Alternatively, one can use a computer to decide on the optimum amount of equipment necessary to fit the patient load. This may require the application of queuing theory [45,46]. A final business application of the microcomputer is that of computer-to-computer communications. This is usually accomplished via the telephone lines by means of a special interfacing device called a modem. With such a device, a physician anywhere in the world can tie into the powerful National Library of Medicine (NLM) computer system. This allows even an isolated physician to use MEDLARS, the computerized medical literature research service offered by the NLM. The NLM also maintains computer data bases on toxiocology, cancer literature, epilepsy, and other subjects. All of these data bases can be queried for a modest fee. Using a modem, one can also tie into commercial data networks for a nominal fee. These networks allow one to order airline tickets from home, or to query large newspaper data bases, such as that of the New York Times. Physicians interested in the stock exchange can now have their microcomputers automatically dial the Dow Jones computer, analyze one’s current portfolio, and

Applications

of a microcomputer

for the general

radiologist

59

hang up automatically. It is also possible to communicate with the computers of other physicians, and send not only X-ray reports, but the images themselves across the country. Once sufficient people have both computers and modems, instant electronic mail will be a reality. Any microcomputer which is attached to a modem is no longer merely a microcomputer. The modem can allow one to communicate with a much more powerful computer, and use its greater power to run certain difficult programs, such as the AI programs mentioned earlier in this paper. In this application. one’s microcomputer becomes an “intelligent” remote terminal for the larger system. Such services are provided by the same commercial data networks mentioned above. MISCELLANEOUS

APPLICATIONS

I will now address the applications of microcomputers to a special class of radiologist: the handicapped radiologist. Though most radiologists are free from major sensory or motor handicaps, the spectre of someday developing them looms over us all, Blindness, stroke, trauma, and other cripplers strike not only radiologists but our family members and friends as well. It is of some comfort, then, that the microcomputer can be used to help rehabilitate these victims. The basic principle here is to interface the computer to one’s remaining senses or muscles so that they can somehow do the job of the missing ones. Many of these applications have been addressed by Kornbluh [47]. Using modems, two deaf persons can communicate over the telephone lines by using their microcomputers as teletypes. A voice recognition and synthesis accessory is currently available for many microcomputers. A deaf person could use this device to convert the sounds he cannot hear on the telephone into printed words on his computer screen. With the aid of such a device, a quadriplegic, a stroke victim, or a child with cerebral palsy could speak commands into a microphone, and thereby drive a wheelchair, feed themselves, or earn a living. A related device can scan printed text. and convert the printed word to the spoken word for a blind person. Most of these applications can be implemented on a microcomputer. The reader who is interested in the plight of the disabled radiologist is directed to an article by Redden [48], describing the American Association for the Advancement of Science (AAAS) Project on the Handicapped in Science. This project has served as an advocacy and information resource for disabled professionals and students of science since 1975. It also publishes the Rrsourcr Dirrctory qf’ Hundicapperl Scientisrs, which is available from the AAAS [49]. SUMMARY A brief overview has been given of the current status of computers in radiology, and an attempt has been made to describe the place of the microcomputer in this dynamic field. Multiple examples have been given of both potential and actual microcomputer applications in radiology. It is hoped that this paper will assist the reader in developing applications for his own needs in radiology. REFERENCES G. Freiherr.

History of computing. The Seeds of‘ Artificial Intelligcwe SUMEX-AIM. National Institutes of Health, Bethesda. MD. (NIH Publication No. 80-2071) (1980). R. M. White. Disk-storage technology. S&m. Am. 243, 138-148 (1980). R. P. Kruger, W. B. Thompson and A. F. Turner. Computer diagnosis of pneumoconiosis. I.E.E.E. Crherncrics 4, 4s-49 ( 1974). G. Wagner. P. Taugu and U. Wolber. Problems of medical diagnosis: a bibliography Mrrh. f@m. Med. 17, 55 74 (1978). P. V. Houdek. The use of microcomputers for a radiotherapy treatment planning. Proc. Sisrk Con/: Compter Applicutions

in Radiology

and Computer/Aided

Analysis

of Radiological

Images.

pp. 52-55

(1979).

V. L. Yu. L. M. Fagan and S. M. Wraith et al., Antimicrobial selection by a computer: a blinded evaluation by infectious diseases experts. J. Am. Med. Assoc. 242. 1279-1282 (1979). C. 1. Henschke. S. J. Hessel and B. J. McNeil, Automated diagnosis in radiology. Itnc~st. Radio/. 14, 195-201 (1979). E. H. Shortliffe, B. G. Buchanan and E. A. Feigenbaum. Knowledge engineering for medical decision making: a review of computer-based clinical decision aids. Proc. I.E.E.E. 67. 120771224 (1979). F. R. Freernon, Computer diagnosis of headache, Headache 8, 49-56 (1968). H. R. Warner, A. F. Toronto, L. G. Veasey and R. Stephenson. A mathematical approach to medical diagnosis. J. ilm. med. Assoc. 177, 177-183 (1961). E. A. Patrick. Introduction to probability and statistics. Decision Anu/ysis in Medicine: Methods und Applicutiorw. pp. 7 8. CRC Press, Boca Raton (1979).

60

MICHAEL L. RICHARDSON

12. G. S. Lodwick, A probabilistic approach to the diagnosis of bone tumors, Radio/. Clin. N. Am. 3. 487-497 (1965). 13. D. G. Fryback and J. R. Thornbury, Evaluation of a computerized Bayesian model for diagnosis of renal cyst vs. normal variant from excretory urogram information, Invest. Radio/. 11. 102-111 (1976). 14. W. J. Wilson, A. W. Templeton, A. H. Turner and G. S. Lodwick, The computer analysis and diagnosis of gastric ulcers, Radiology 85, 1064-1073 (1965). 15. A. W. Templeton, C. Jansen, J. L. Lehr and R. Hufft, Solitary pulmonary lesions: computer-aided differential diagnosis and evaluation of mathematical methods, Radiology 89, 605-613 (1967). 16. A. F. Toronto, L. G. Veasy and H. R. Warner, Evaluation of a computer program for diagnosis of congenital heart disease, Prog. Cardiouasc. Dis. 5, 362-377 (1963). 17. B. J. McNeil and H. Sherman, Example: Bayesian calculations for the determination of the etiology of pleuritic chest pain in young adults in a teaching hospital, Comp. Biomed. Res. 11, 187-194 (1978). 18. J. P. Whalen, Radiology of the abdomen: impact of new imaging methods, Am. J. Roentg. 133, 585618 (1979). of acid base disorders. J. C/in. Invest. 48, 1689-1696 (1969). 19. H. L. Bleich, Computer evaluation What is artificial intelligence?, Computers in Medicine: Appbtions for Artificial Intelligence 20. E. A. Feigenbaum, Techniques/Continuing Education Tutorial, pp. 3-6. Stanford University School of Medicine, CA (1980). 21. H. E. Pople, Heuristic methods for imposing structure on ill-structured problems: the structuring of medical diagnostics. Decisions Systems Laboratory University of Pittsburgh (1980). Unpublished technical report. 22. L. Frenzel, The personal computer-last chance for CAI? Byte 15, 8696 (1980). 23. S. L. Gammill, A Programmed Introduction to Upper Gastrointestinal Radiology. Little Brown, Boston (1977). Intelligent computer-aided instruction for medical diagnosis. PFOC. 24. W. J. Clancey, E. H. Shortliffe and B. G. Buchanan, 3rd Ann. Symp. on Comp. Applic. to Med. Care Con& pp. 1755183 (1979). Reading (1980). 25. J. E. Randall. Microcomputers and Physiological Simulation. Addison-Wesley, insiruction in medical education, May 19?4_April i978. National Library of Medicine, 26. Computer aided and prdgrammed Bethesda. MD (1978). (NLM literature search no. 78-l 1). 27. C. J. Tegtmeyer; T. 6. ‘Keats, E. W. Pullen and J. Langman, The teaching of roentgen anatomy to medical students: a self-instructional approach, J. Med. Educ. 49, 455-456 (1974). 28. J. A. Crocco, J. J. Rooney, H. S. Diamond, C. M. Plotz and M. Weiner, A computer-assisted instruction course in the diagnosis and treatment of respiratory diseases, Am. Rev. Respir. Dis. 111, 299-305 (1975). 29. H. S. Diamond, M. Weiner and C. M. Plotz. A computer-assisted instructional course in diagnosis and treatment of the rheumatic diseases, Arthritis Rheum. 17, 1049-1055 (1974). 30. C. J. D. Smith, Use of a programmed Br. J. Med. Educ. 9,27T33 (1975). reinforcement exercise in teaching of embryology, a pathophysiological 31. A. H. Goroll, G. 0. Barnett, J. Bowie and P. Prather. Teaching differential diagnosis by computer: approach, J. Med. Educ. 52, 153-154 (1977). 32. A. A. Steele. P. J. Davis, E. P. Hoffer and K. T. Famiglietti, A computer-assisted instruction (CAI) program in diseases of the thyroid gland (THYROID), Comput. Biomed. Res. Il. 133-146 (1978). 33. 0. Elgenmark, The normal development of the ossific centers during infancy and childhood; clinical, roentgenologic, and statistical study, Acta paediat. 33, (Suppl. l), 1-79 (1946). 34. A. F. Roche, H. Wainer and D. Thissen. The RWT method for the prediction of adult stature, Pediatrics 56, 102&1033 (1975). 35. L. Poole and M. Borchers. Some Common Basic Programs, 2nd edn. Adam Osborne & Associates. Berkeley (1978). What’s that? Semin. Nucl. Med. 8, 316-323 (1978). 36. R. S. Bell. Efficacv.. on the decision theoretic apprdach ;o medical images, Semin. Nucl. Med. 8, 307-315 37. R. F. Wagner, Some perspectives (1978). 38. C. E. Metz, Basic principles of ROC analysis, Semin. Nucl. Med. 8, 283-298 (1978). 39. L. B. Lusted, General problems in medical decision making with comments on ROC analysis. Semin. Nucl. Med. 8, 299-306 (1978). to clinical decision making, Semin. Nucl. Med. 8. 273-282 (1978). 40. D. D. Patton, Introduction 41. G. W. Hamilton. G. B. Trobauch. J. L. Ritchie. K. L. Gould. T. A. DeRouen and D. L. Williams. Mvocardial imagine with 2”‘thallium: an analysis of clinical usefulness based on Bayes’ theorem, Semin. Nucl. Med. 8, 3581364 (1978). y y 42. A. Watson, A simplified theory of video graphics: part 1, Byte 5, 18&189 (1980). 43. G. T. Herman and C. G. Coin, The use of three-dimensional computer display in the study of disk disease. J. Comput. assist. Tomogr. 4, 564-567 (1980). 44. L. L. Rose and M. H. Gotterer, Computerized patient scheduling in a clinic, Proc. Second Ann. Symp. Computer Application in Medical Care, pp. 478-485 (1978). of Queuing theory. PFOC. Fourth Ann. Symp. Computer Applications in Medical Care, pp. 45. J. L. Ford, Two applications 169&1693 (1980). MA (1977). 46. R. B. Coats and A. Parkin, Computer Models in the Social Sciences. Winthrob, Cambridge. Application of technology to handicapped individuals: process and issues, Proc. I.E.E.E. Computer Society 47. M. Kornbluh, Workshop on the Application of Personal Computing to Aid the Handicapped, pp. 5-10 (1980). 48. M. R. Redden, Disabled scientists: who are they and what can they tell us?, Proc. I.E.E.E. Computer Society Workshop on the Application of Personal Computing to Aid the Handicapped, pp. 75-78 (1980). 49. AAAS Project on the Handicapped in Science; 1776 Massachusetts Avenue, N.W., Washington, DC.

About the Author-MICHAEL L. RICHARDSON received the B.S. degree in physics from Texas A & M University in 1972 and the M.D. degree from Bavlor Colleae of Medicine in 1975. Following internshio. Dr Richards& spent 2 years as a staff internist a; Vandenberg AFB, CA. and then completed his specialty training in diagnostic radiology at David Grant USAF Medical Center. Travis AFB. CA in 1981. He is currently a staff radiologist at the Mather AFB hospital and a Clinical Instructor in Radiology at the University of California, Davis, Medical Center. Besides computerized radiology, he is also very interested in cross-country skiing and the banjo.