YT YouTube2Text

YouTube2Text → Transcripts → Najgroźniejszy błąd w historii radioterapii

Najgroźniejszy błąd w historii radioterapii — Transcript

by OdwróconaLekcja · 6,105 words · 1,001 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00The patient felt a rush of heat, as he
0:02later recalled, as if someone had
0:04poured hot coffee on his back. The
0:06machine stopped working and displayed
0:08the message "malfunction 54." At the
0:10same time, the operator received
0:12information that the device had
0:14delivered a fraction of the required
0:16dose. Following procedure, she pressed
0:19"P" for "proceed" to continue. She
0:22didn’t know that in less than a
0:24second, the patient had received a
0:25radiation dose over 100 times higher
0:27than planned. And this was even before
0:30she pressed "B." We are talking about a
0:32radiotherapy device intended to destroy
0:34cancer cells. This story became one of
0:37the most infamous disasters at the
0:39intersection of medicine, physics,
0:40radiation, and computer science. My
0:42name is Karol Bielski. This is "
0:44Odwrócona Lekcja"(Inverted Lesson). If
0:46you feel like supporting us with a
0:47subscription, we would be very pleased.
0:49Though, of course, the decision is
0:51yours. Let’s get started. Chapter one
0:53. A new generation machine. In the
1:06early 70s, the Canadian state-owned
1:08company Atomic Energy of Canada Limited
1:10(AECL) collaborated with the French
1:12company CGR to produce radiotherapy
1:14equipment. The fruit of this
1:17cooperation was two modern machines for
1:19those times. The Therac-6 and the
1:22Therac-20. Both devices were equipped
1:25with a PDP-11 computer, but its role
1:27was marginal. It helped the operator
1:31enter parameters and simplified
1:32operation, while the machine’s most
1:34important functions could run
1:35independently of this computer. Safety
1:39was overseen by mechanical interlocks
1:40and separate electrical circuits. If
1:44individual parts of the device were not
1:46in the correct position during patient
1:48irradiation, the machine would not
1:49allow it. It was physically impossible
1:53to deliver the electron beam. In the
1:55mid-70s, AECL began work on the next
1:58generation of accelerator. The new
2:02design used a dual-pass electron system
2:04, allowing the device to achieve high
2:06energy while taking up less space. The
2:11first prototype was created in 1976,
2:14and at the end of 1982, the fully
2:16computerized version, the Therac-25,
2:18hit the market. Before we go any
2:23further, during this past vacation, I
2:25was resting while my money kept working
2:27, all thanks to Trade Republic. The
2:30partner of today's episode. Not long
2:33ago, I kept my money in several
2:34different accounts without a specific
2:36plan and, frankly, without much profit.
2:39Then I discovered Trade Republic and a
2:41lot has changed. It is a licensed
2:44German bank trusted by over 10 million
2:46customers in Europe. In one simple app,
2:50I save, pay, and even—albeit timidly
2:53for now—invest. What appealed to me
2:56most was the interest rate on cash.
2:58Trade Republic currently offers 6%per
3:00annum on funds in the account. Interest
3:03is calculated daily and paid out
3:05monthly, and the money is available at
3:07any time without locking it up. Unlike
3:10a fixed-term deposit, new customers
3:12receive 6%interest on balances up to
3:14200,000 PLN, and amounts above this
3:16limit are subject to interest equal to
3:19the NBP deposit rate. Here is an
3:21example. If you keep 10,000 PLN in the
3:23account, then at 6%per annum, this
3:25gives you 600 PLN gross per year.
3:28that's fifty gross per month, and you
3:30can see it all clearly right there in
3:32the app. These transparent rules are
3:35simply excellent. You open an account,
3:37turn on the interest in the app, and
3:39watch your earnings grow. With every
3:41interest payout, the withholding tax is
3:43automatically deducted, so you don't
3:45even have to worry about that. It's
3:48also worth mentioning that when paying
3:51abroad, there are no additional bank
3:53fees for currency conversion, and Trade
3:55Republic charges no fees for ATM
3:57withdrawals over 200 PLN. This was
4:00really useful for me during my holiday
4:02travels. Instead of looking for
4:04exchange offices, you just pay with
4:05your Trade Republic card and don't
4:07worry about anything else. Opening an
4:09account takes a few minutes and is done
4:11entirely online. Scan the QR code
4:13visible on the screen, download the app
4:16, and start saving. Starting today with
4:18Trade Republic. And now, let's get back
4:20to the material. In the first mode, the
4:22device directed an electron beam at the
4:25patient with energy from 5 to 25 MeV.
4:29Such a beam was primarily effective for
4:30treating lesions located relatively
4:32shallowly in the patient's body. The
4:35second mode was used for treating
4:36tumors located deeper. The Therac-25
4:39then accelerated electrons to an energy
4:41of 25 megaelectronvolts or more and
4:43directed them at a metal target. This
4:47collision led to the creation of X-ray
4:49radiation. X-ray radiation, which then
4:52passed through a special filter. To
4:56generate X-rays, an electron stream
4:58with an intensity about 100 times
5:00greater than during direct electron
5:02therapy was required. Inside the head,
5:06there was a rotating mechanism that
5:08positioned the elements we mentioned
5:10earlier, the target and the filter, in
5:12the path of the beam. The first
5:14position held the scanning magnets,
5:16used during electron therapy, the
5:17lighter one, let's say. The second held
5:21the target, the filter, and the chamber
5:23measuring the X-rays. The third
5:26position was used for aiming at the
5:27patient's body. In that position, a
5:31steel mirror was in the beam's path,
5:33and a simple light showed the operator
5:35the spot to be irradiated. In this last
5:39position, no dose-measuring sensor was
5:41installed. After all, the designers
5:44didn't anticipate that the accelerator
5:46could emit radiation then. I remind you
5:49, there was supposed to be only a light
5:51, you know, like a laser. A laser, not
5:54a light, which would simply show on the
5:55patient's body where we're going to
5:57shoot. And then the configuration would
6:02change to the first or second,
6:03depending on whether we want that
6:05lighter dose of electrons, or that
6:06X-ray radiation. If the device had
6:13triggered the strong beam intended for
6:15X-ray production, and the target and
6:17filter weren't in its path, the patient
6:19would have been hit by a concentrated
6:21stream of electrons without the
6:22elements meant to convert, weaken, and
6:24spread it. In older devices,
6:30independent mechanical and electrical
6:31interlocks protected against such a
6:33situation. When designing the Therac-25
6:37, AECL abandoned such old-fashioned and
6:39outdated safeguards. The manufacturer
6:44decided that a computer could
6:45simultaneously operate the device,
6:47check the position of the elements, and
6:49prevent the beam from starting in the
6:51wrong configuration. So, the
6:53responsibility was shifted from
6:55physical mechanical safeguards to
6:57digital locks. managed by a computer.
6:59The control program for the Therac-25
7:01was written in assembly for the PDP-11
7:04computer. It was developed by a single
7:06person. The entire software for a
7:09device that shoots radiation, electrons
7:11, at patients. One person, and they
7:16used code snippets that came from even
7:18older models. Documentation was sparse,
7:21and separate tests for the software
7:23itself were very limited in scope. The
7:27manufacturer later claimed during
7:29hearings that the system had undergone
7:31about 2,700 hours of testing. It turned
7:34out, however, that this number
7:36represented 2,700 hours of device usage
7:38. So, the tests were, well, they were
7:40only on the bodies of patients. In 1983
7:43, AECL conducted a safety analysis of
7:46the Therac-25. It assumed that
7:50simulator tests and prior usage had
7:51limited the number of programming
7:53errors. Perfect. Potential problems
7:57that might have remained in the code
7:59were not considered. The analysis
8:02covered possible computer failures, but
8:04it viewed them primarily as the result
8:06of hardware damage or random
8:08interference. The possibility that the
8:11program itself could have flaws was
8:13ignored. I got a bit confused. It was
8:16simply assumed that since everything
8:18worked for 2,000 hours, it would keep
8:20working, so why dig deeper? Ultimately,
8:2311 Therac-25 units were installed in
8:25the United States and Canada. The
8:29devices began treating patients and
8:30worked wonderfully for the first few
8:32years, and no officially confirmed
8:34problems were reported. Everything
8:38turns upside down on June 3, 1985, in
8:41Marietta, Georgia. After the beam was
8:45activated, the patient felt a wave of
8:47heat. When the operator entered the
8:50room, the woman said the machine had
8:51burned her. Please don't worry, that's
8:54impossible. Chapter two. That's
8:57impossible. The 61-year-old woman had
9:08previously undergone surgery to remove
9:10a breast tumor. Radiotherapy was
9:12intended to destroy cancer cells that
9:13might have remained in the surrounding
9:15lymph nodes. On June 3, '85, she was
9:19being prepared for another radiation
9:21treatment in the collarbone area with a
9:2310 MeV electron beam. Soon, redness and
9:28swelling appeared in the irradiated
9:30area. The pain grew, the shoulder
9:32stiffened, and muscles began to
9:34contract in such painful spasms. The
9:37woman ends up in the hospital, but
9:39doctors still referred her for further
9:40treatments at the center. Initially, it
9:44was suspected that the changes resulted
9:46from the progression of the disease or
9:48some unusual reaction to properly
9:49performed radiotherapy. However,
9:52medical physicist Tim Steel, who worked
9:54at the center, began to suspect a
9:56malfunction. He contacted AECL and
9:59asked if the Therac-25 could trigger an
10:01electron beam without the scanning
10:03magnets, which would, as it were,
10:05spread or disperse it across the
10:06treatment field. Such a situation, my
10:11dear sir, is. impossible. About two
10:14weeks after the procedure, redness
10:16appeared on the patient's back,
10:17corresponding to the lesions visible on
10:19the front of her body. The skin in the
10:21irradiated area begins to peel. The arm
10:23remains immobile. The woman suffers
10:26from constant pain. The nature of the
10:29injuries indicated severe radiation
10:31burns, but no one could explain how the
10:33Therac-25 could have caused them.
10:37Physicist Ken Kenstone later estimated
10:39that the patient may have received a
10:41dose of 15,000 to 20,000. At that time,
10:45they were measuring in rads, 15,000 to
10:4720,000 rads. A typical single
10:50therapeutic dose is about 200 rads. A
10:55precise reconstruction of the event
10:56proved impossible because the treatment
10:58parameter printing function was
11:00disabled at the time. There was no
11:04record left of the device settings.
11:07Meaning, no one knew exactly how the
11:09operator had set everything up to get
11:11that result. The woman's injuries were
11:13irreversible. Her breast was removed,
11:15she lost the ability to move her arm,
11:17and she was left with chronic pain. The
11:20other facilities—I remind you there
11:22were 11 of them—using the Therac-25
11:24were not informed about anything. The
11:28matter was also not reported to the
11:29American Food and Drug Administration.
11:32Regulations at the time imposed a duty
11:34to report malfunctions and injuries.
11:37Primarily on equipment manufacturers
11:39and importers. Hospitals and employees
11:42had no such obligation. At least not
11:44yet, because those standards were
11:46changed later. It is seven weeks after
11:48the accident in Marietta. We are moving
11:50to Canada now. On July 26, 1985, a 40-
11:53year-old patient arrives at the Ontario
11:56Cancer Foundation in Hamilton for her
11:5824th cervical cancer radiotherapy
12:00treatment. The operator starts the
12:05device, but after 5 seconds, the Therac
12:07-25 stops and displays an "H-tilt"
12:09error message, indicating beam
12:11asymmetry. Meaning something was wrong
12:14with the beam. However, the measurement
12:16system indicated that the patient was
12:18receiving no dose at all. The device
12:21classified the event as a pause, a
12:23break in operation, after which the
12:25treatment could be continued without
12:27re-entering parameters. So we have
12:31everything saved there, just press "P",
12:33no need to set anything up. You can
12:37fire electrons again. So the operator
12:39presses "P"—Proceed. Proceed. What's
12:43going on? It's not working again. "P"
12:45—The procedure is repeated. Damn it,
12:47it's jammed again. After the fifth
12:50attempt, the fifth attempt, the device
12:52went into a full halt state, as if it
12:53had frozen. For the staff, this was not
12:56an exceptional event. The Therac-25
12:59frequently interrupted work and
13:01displayed error messages about
13:02insufficient dose or incorrect beam
13:04setting. Usually, it was not associated
13:07with any danger to the patient. In some
13:09facilities, similar errors occurred
13:11even a dozen or several dozen times a
13:13day. In any case, five attempts were
13:16made. After the treatment, the patient
13:19complained of a burning sensation and a
13:21feeling resembling an electric shock in
13:23the hip area. Despite this, on the same
13:28day, six more people were irradiated
13:30using the same device. Three days later
13:33, the woman returns to the center. The
13:35pain would not subside. Swelling
13:37appears at the site of the treatment.
13:39The Therac-25 was taken out of service.
13:42The patient was admitted to the
13:43hospital with a suspected radiation
13:45overdose. AECL and Canadian radiation
13:48protection authorities were notified.
13:51Later estimates indicated the woman may
13:53have received 13,000 to 17,000 rads,
13:55though it is unknown how many of those
13:57five attempts were actual hits. Maybe
13:59all of them, or maybe just one, a very
14:01powerful one—it’s unknown. The
14:04patient died on November 3, 1985,
14:06though the autopsy stated it was due to
14:08an aggressive tumor, so the cause of
14:10death was cancer, not radiation.
14:16However, the injuries caused by the
14:17Therac were so severe that if she had
14:19survived, if the cancer hadn't killed
14:21her, she likely would have needed a
14:23total hip replacement. AECL engineers
14:27were unable to reproduce the failure.
14:29Suspicion fell on one of the
14:31microswitches that informed the
14:32computer about the position of the
14:34rotating mechanism. You know, it’s
14:35about that rotating mechanism that,
14:37when necessary, positioned the shield
14:39and the filter, or a mirror, or nothing
14:40at all. It was concluded that a
14:43momentary sensor failure could have
14:44sent incorrect information to the
14:46system and allowed the device to
14:47operate in the wrong configuration.
14:50AECL modified the sensor system and
14:52introduced additional software checks
14:54for the mechanism's position. The
14:57manufacturer claimed these changes
14:58reduced the risk of failure by at least
15:00five orders of magnitude. Something
15:03like 100,000 times less chance of a
15:05failure. Canadian experts recommended
15:08installing an independent
15:10hardware-based physical system to
15:11monitor the position of the rotating
15:13mechanism. They demanded that errors
15:17related to dose measurement immediately
15:19stop the treatment and prevent it from
15:21being resumed by just pressing a key.
15:23Meaning if something hangs, if
15:25something happens, if a message pops up
15:27, it shouldn't be enough to just press
15:29'P' to resume work. AECL did not
15:31implement all of those recommendations
15:33at the time. The number of possible
15:35retries was reduced from five to three
15:37before a system hang, but no
15:38independent additional lock for the
15:40mechanism's position was installed. A
15:44letter sent to users informed them of
15:45the issues with the device and
15:47recommended visually checking its
15:48settings. It did not mention that one
15:52of the patients had been seriously
15:53injured. After the modifications were
15:56made, the centers received permission
15:58to return to normal use of the Therac-
16:0025. Great. The manufacturer still
16:03didn't know what had happened. Chapter
16:05three: X and E. One of the devices
16:17modified after the accident in Hamilton
16:19was a Therac-25 operating at the Yakima
16:21Valley Memorial Hospital in Washington
16:23state. In December 1985, a patient
16:25treated there noticed red, stripe-like
16:28marks on her right hip. The staff tried
16:30to determine the cause. They suspected
16:33the elements used to shape the
16:34radiation field. They also considered
16:37reactions to chemotherapy. However,
16:39that should have appeared in other
16:40radiated areas as well and likely
16:42wouldn't have formed parallel stripes.
16:45When the staff learned that the woman
16:47used a heating pad at home, suspicion
16:49fell on the internal wires, where I
16:51think warm water flows. The pad was
16:53X-rayed, but the layout of the wires
16:55did not match the marks on the
16:57patient's body. So, the woman continued
16:59her radiotherapy. It ended on January 6
17:02, 1986. The staff still didn't know
17:05what caused the stripes, or these
17:06unusual injuries. On January 31,
17:08hospital staff described the case in a
17:11letter to AECL. They spoke by phone
17:13with a technical support manager. The
17:16response arrived on February 24, after
17:18almost a month, and AECL stated that
17:20the bodily injuries could not have been
17:22caused by either a Therac failure or
17:24operator error. They just couldn't
17:27imagine how such stripes could have
17:29been formed. They attached two pages of
17:33technical arguments as proof, claiming
17:35that such an operation, such a failure,
17:37was impossible. Furthermore, they
17:41argued that similar injuries had not
17:43been observed in other patients. Except
17:46we already know that while the injuries
17:48were not the same, they had been
17:49observed; I remind you of the cases
17:51from Kenstone and Hamilton. However,
17:52individual hospitals do not know that
17:54an accident is happening in another
17:56city somewhere. Each center seemed to
17:59only see its own piece of the problem,
18:01and this problem was much, much wider.
18:03The staff in Yakima trusted the
18:05manufacturer. The Therac had undergone
18:06modifications, and these were supposed
18:08to radically increase safety. I remind
18:10you, 100,000 times. Ultimately, the
18:13cause of the injuries was labeled as
18:15unknown. Less than a month after the
18:17response sent to Yakima, on March 21,
18:191986, a patient at the East Texas
18:21Cancer Center in Tyler arrived for his
18:24ninth radiotherapy treatment. He had
18:26recently had a tumor removed from his
18:27back, and the planned treatment was
18:29intended to destroy the remaining
18:31cancer cells. The doctor prescribed an
18:33electron beam with an energy of 22 MeV
18:35and a dose of 180 rads. The man lay
18:37down on his stomach. The operator left
18:39the room and closed the heavy door
18:41behind her. It is the kind of door that
18:43is supposed to protect her from
18:44radiation. After all, she spends the
18:46whole day somewhere near this machine.
18:48That day, the video monitor was
18:49disconnected, and the intercom was not
18:51working. They couldn't hear each other.
18:53That is, the female operator in one
18:55room and the patient behind the heavy
18:57door. The woman had been working with
19:00the Therac for a long time and operated
19:02the machinery and its keyboard very
19:03efficiently. While entering the
19:06parameters, out of habit, she chose the
19:08X-ray mode marked with the letter X,
19:09but she was a professional. Almost
19:12immediately, she noticed the mistake.
19:14She moved the cursor back and changed X
19:16from X to E, that is, from X-ray
19:18radiation to this electron radiation,
19:20those pure electrons. The other
19:23parameters matched, everything was
19:25correct, the operator confirmed it, and
19:27the computer displayed information that
19:29the device was ready for operation. So,
19:31she pressed the key to activate the
19:33beam. The patient felt a sudden strike
19:36and heat. He also heard a loud buzzing
19:39coming from the device. The Therac
19:41interrupted its operation; the screen
19:43displayed the message "Malfunction 54."
19:45The system indicated that instead of
19:47the planned dose, it had delivered a
19:49small fraction of it. The operator
19:51therefore followed the standard
19:53procedure, what the manufacturer
19:55recommended, and pressed "P" to proceed
19:57. At the same moment, the patient was
19:59trying to get up from the table. Well,
20:01they couldn't hear each other. He
20:03didn't know that she wanted to continue
20:04. Chapter four. Malfunction 54. The
20:16Therac activated the beam. The man felt
20:18a strike, this time resembling an
20:20electric shock running through his
20:21shoulder. He was already getting up. He
20:23was shot with the beam while he was in
20:25that position, you know, already
20:26half-sitting. He had the impression as
20:28if his hand had detached from the rest
20:30of his body. He got off the table,
20:32approached the door, and started
20:33banging on it. The operator immediately
20:35stopped the procedure and let him into
20:37the room next door, into the control
20:38room. The doctor noticed intense
20:40redness, a classic symptom, but assumed
20:42the man might have been shocked by
20:44electricity. Go home, and if anything
20:48happens, you know where to find us. A
20:51medical physicist working at the center
20:53, Fritz Heiger, was also called in. He
20:56checked the Therac’s calibration,
20:57finding nothing, no irregularities at
20:59all. Curiously enough, no one really
21:02knew what the "Malfunction 54" message
21:04meant. The next day, two AECL engineers
21:09arrived. One from the local branch, the
21:11other from headquarters in Canada. They
21:13tested the device all day, but they
21:15couldn't trigger the error. One of the
21:18engineers assured them that the Therac
21:2025 could not deliver an overdose. When
21:24Heiger asked about other cases, he was
21:26told that AECL knew of nothing similar.
21:30The manufacturer suggested the patient
21:32had indeed been shocked by electricity,
21:34so an independent engineering firm
21:36examined the installation but found no
21:38electrical fault. On April 7, the
21:40Therac returns to service. Four days
21:42later, on the 11th, another patient is
21:44treated. The same operator enters the
21:46treatment parameters. Damn, X instead
21:49of E again. Will I ever learn? She
21:51instantly notices the mistake, hits the
21:53correct key, changes the mode to
21:55electron, and quickly confirms the
21:57remaining data. The computer stated
21:59that the beam was ready. The operator
22:01activates the device. After a few
22:03seconds, the Therac stops with a buzz,
22:05displaying "Malfunction 54" on the
22:07screen. This time, the intercom was
22:10working. The operator heard moaning and
22:12immediately ran into the room. The man
22:14was trying to tear off the tape holding
22:16his head in place. He said he felt fire
22:17on the side of his face. I don’t
22:19think I mentioned that it was aimed
22:20right at his face. He later described
22:22that something had hit him. He saw a
22:25flash and heard a sound like eggs
22:27frying. This time, the device was
22:30immediately taken out of service. The
22:33operator remembered the sequence of
22:34actions perfectly. Accidentally
22:37selecting X, returning the cursor to
22:38the previous field, changing to E,
22:40confirming, and triggering the beam.
22:43Heiger tried to repeat the sequence,
22:45but nothing happened. The Therac was
22:48working correctly, and then it hit him:
22:50the speed. A physicist who didn’t
22:53operate the machine daily wasn’t as
22:55efficient; you know, he was like a
22:57woodpecker with a keyboard. He wasn't
22:59giving commands to the machine any
23:01slower than the experienced operator.
23:03He had to practice, and he actually
23:05learned to enter the corrections just
23:07as fast as she did. Once he succeeded,
23:10he immediately saw "Malfunction 54" on
23:12the screen. Ten months after the first
23:16known episode of the Therac 25, the
23:18failure leading to a massive overdose
23:20was finally reproduced. I haven't found
23:25information anywhere on how many people
23:27in total were treated on all the
23:28machines during that time. Heiger began
23:31measuring the dose emitted during the
23:33failure. The first result was 804 rads,
23:36but he quickly realized the beam was so
23:38intense that it was overloading the
23:40measuring equipment. He adjusted the
23:44sensors and received a result exceeding
23:464,000 rads. So, it isn't the
23:48manufacturer discovering what really
23:50happened, because the company still
23:51cannot replicate the error. It was
23:54Heager who taught them that sequences
23:56must be performed quickly. Only then
23:59were the AECL engineers also able to
24:01trigger the malfunction. Measurements
24:03conducted by the manufacturer showed
24:04that in the center of the field, the
24:06device could deliver as much as 25,000
24:08rads. The exact doses received by the
24:10patients remain unknown, because they
24:12varied wildly during subsequent
24:14simulations, so it's not entirely clear
24:16who got what. But it is known that all
24:19the victims received far too much.
24:22Chapter five. 8 seconds. You might ask,
24:34what was this actually about? Where did
24:36the problem lie? The Therac accelerates
24:38electrons in both modes, right? The
24:41catastrophe occurred when the Therac
24:43combined elements of one operation with
24:44the settings of another. After
24:49initially selecting X, the computer
24:51began preparing the device to produce
24:53X-rays, meaning that stronger beam.
24:57Setting the magnets took about 8
24:59seconds before everything rotated into
25:00place. If, within that time window, in
25:05the middle of those 8 seconds, the
25:07operator changed the settings to e, the
25:09target and filter would move out of the
25:10beam's path because they aren't
25:12supposed to be there, but the energy
25:14remains. So the accelerator behaves
25:20according to the original X setting,
25:21but the safety measures get out of its
25:23way. The Therac is, as it were, in two
25:28different states at once. A
25:30Schrödinger's Therac. The operator saw
25:32electron therapy on the screen. The
25:35rotating mechanism was also in the
25:37position intended for that type of
25:39treatment, but the accelerator was
25:41blasting the massive stream needed for
25:42X-rays. But energy isn't everything.
25:46The problem was also the concentration
25:48of the beam, which in these two
25:50settings was not scattered and, like an
25:52electron blade, was focused on a small
25:54area. It was cutting through the
25:56patient's body with electrons, causing
25:58terrible destruction. I don't know if I
26:00can put it that way. I think you
26:02understand for the sake of imagination.
26:05A huge number of electrons also
26:06overloaded the ionization chambers,
26:08which measured the dose. The sensors
26:11stopped working properly and informed
26:13the computer that the patient was
26:14receiving a fraction. That's why the
26:17Therac displayed "malfunction 54," but
26:19simultaneously suggested that there had
26:21been an underdose. So the operator had
26:24the option to press "Set" again,
26:28injuring the patient. According to the
26:31AECL engineer, the sound resembling
26:33frying eggs that he heard came
26:34precisely from those damaged and
26:36overloaded ionization chambers. And
26:40I'll be honest, when we first read that
26:42, I thought it might be about the
26:43denaturing of proteins in the body.
26:46Fortunately, no. So the whole problem
26:49was contained within eight seconds.
26:52Several procedures were running
26:53simultaneously and using the same
26:55information. The result depended on
26:57which one would be first. A race of
26:59sorts. That is exactly what is called a
27:01race condition. If the operator worked
27:03a bit slower, wasn't quite as efficient
27:05, the program would register the
27:07correction and correctly prepare
27:08everything for work. The error,
27:11therefore, had a very perverse
27:12characteristic. It was most easily
27:15triggered by the person who knew the
27:17device best and operated it the fastest
27:18. The second patient, Styler, died on
27:21May 1st, less than three weeks after
27:23the accident. The autopsy revealed
27:25acute radiation damage to the right
27:27temporal lobe and the brain stem. brain
27:30. Besides, the first patient's
27:31condition was also deteriorating. He
27:33lost control of his left arm. He
27:35suffered paralysis of both legs, one
27:37vocal cord, and his diaphragm. He died
27:39five months after the accident. Chapter
27:42six. Tape. Four days later, AECL sent
27:53Therac users instructions for a
27:55temporary fix to the problem. What you
27:57are about to hear now is not a joke. We
27:59have confirmed this through several
28:01sources. You had to take a screwdriver
28:04to the keyboard, remove the cursor-up
28:06key, and disable the contacts
28:08underneath with electrical tape. If the
28:12operator made a mistake while entering
28:14parameters, they had to reset the
28:16device and type everything in from the
28:17start. The primary protection against a
28:21fatal software bug thus became a
28:22ripped-out button and a piece of tape.
28:26The FDA deemed this reaction
28:27insufficient. Unsurprisingly, the
28:30letter to users did not explain the
28:32nature of the defect or what kind of
28:34danger it posed. On May 2, 1986, the
28:38FDA declared the Therac a defective
28:40device and ordered a corrective action
28:42plan. It was to include software
28:46changes, additional hardware safety
28:48measures, a redesign of the dose
28:50measurement system, and replacing
28:52cryptic error codes. With something
28:55clearer, something understandable. On
28:57January 17, 1987, another accident
28:59occurred at the Yakima Valley Memorial
29:02Hospital. The patient was to receive
29:05two small control exposures of four and
29:07three rads. A film was placed under his
29:10body, which was meant to record the
29:11radiation field. This allowed the staff
29:14to check if the device was aiming at
29:16the planned spot. After taking the
29:18images, the patient was supposed to
29:20receive another 79 rads of X-ray
29:22radiation. So that first shot was just
29:24for control. The operator delivers the
29:26first four rads. The Therac goes into a
29:28pause state. The field size changes
29:29automatically and it delivers another
29:31three rads. Then the operator enters
29:34the room to remove the film and
29:35position the patient. Using the manual
29:38controller, he rotated the internal
29:40mechanisms of the Therac to the aiming
29:41position. At that moment, a mirror was
29:44in the path of the beam. It reflected
29:46ordinary light, which showed the
29:48location of the future radiation site
29:50on the body. After finishing the setup,
29:53the operator commanded the machine to
29:54prepare for actual treatment. He forgot
29:58to remove the film from under the
29:59patient, which will be important in
30:01this story. On the screen, a message
30:02appeared informing that the beam was
30:04ready. The operator starts the device.
30:07The Therac begins operation, but the
30:08system shows neither the dose nor the
30:10rate of delivery. After five or six
30:13seconds, the machine stops working. The
30:15device has gone into a pause state. So
30:18the operator presses "proceed." The
30:21Therac activates the beam for the
30:22second time. The machine stops again.
30:25This time, a "flatness" message appears
30:27, indicating a problem with the uniform
30:29distribution of radiation. The operator
30:31heard the patient's voice over the
30:33intercom, but didn't understand it. He
30:35entered the room. The man said he felt
30:37a strong burning sensation in his chest
30:39. The screen indicated that the patient
30:41had received a total of seven rads.
30:43Exactly the sum of those earlier
30:44control exposures. The system did not
30:48record the radiation emitted later.
30:51That same day, burns covering the
30:53entire treatment area appeared on the
30:54patient's skin. After four days, the
30:57redness had arranged itself into
30:59stripes. An identical pattern had
31:00already been observed a year earlier.
31:03At the time, if you recall, the cause
31:05was labeled as unknown. AECL begins an
31:08investigation. However, it was
31:11suspected that the beam might have been
31:13fired while the turntable was still in
31:15the setup position, and a forgotten
31:17film came in handy. A hospital
31:21physicist placed another film under the
31:23device, left the turntable in the setup
31:25position, and ran the Therac with X-ray
31:27parameters. He obtained a mark
31:31resembling a striped pattern. After a
31:35week of checking the equipment, AECL
31:37admitted that the machine's malfunction
31:39was not caused by a mechanical failure.
31:43They then began analyzing the software
31:45and found more errors. The program
31:48repeatedly checked whether the device's
31:50components were in the correct position
31:52. It used a variable for this, called
31:54class 3. A non-zero value meant the
31:57mechanism's position needed to be
31:59checked. Zero allowed the check to be
32:01bypassed, assuming everything was fine.
32:03Forgive me for saying this, but you
32:05will soon understand what it's about.
32:07The programmer wanted the variable to
32:09remain non-zero during treatment
32:11preparation. Instead of assigning it a
32:14constant value, like one, the program
32:17incremented it during every test: 1, 2,
32:203, 4. The variable occupied 1 byte of
32:24memory. It could therefore store
32:27numbers between 0 and 255. Once the
32:31device had been running long enough,
32:34upon reaching 255, further incrementing
32:39reset the counter to zero. Every 256th
32:47pass through this part of the program,
32:49the variable briefly displayed zero.
32:52The system skipped the position check,
32:54and during the accident, the operator
32:56pressed "set." At exactly that moment,
32:59the Therac allowed the accelerator to
33:01start, even though the turntable was in
33:03the setup position. Thus, the mirror
33:06remained in the path of the beam. The
33:08target, the filter, and the electron
33:10scattering system were missing. The
33:12ionization chamber that measures the
33:14dose was also not in that position. The
33:16designers, after all, did not
33:18anticipate that one could irradiate
33:19someone then. The Therac launched a
33:22powerful 25 megaelectron-volt stream.
33:24The beam was scattered and deflected by
33:26the steel mirror, then headed straight
33:28for the patient. Later measurements
33:30indicated that a single activation
33:32could deliver 4,000 to 5,000 rads. The
33:36operator pressed "P" and fired the beam
33:38again. The patient could have received
33:40a total of 8,000 to 10,000. Let me
33:43remind you, a total of 86 had been
33:45planned. The patient died in '87 due to
33:48complications. He had previously
33:51suffered from a terminal form of cancer
33:52, but his loved ones claimed the
33:54failure shortened his life and
33:56subjected him to horrific pain until
33:57the very end. The accident forced
34:00personnel to investigate a patient
34:02treated in Yakima in December '85. The
34:06woman had chronic ulcers and tissue
34:08necrosis, suffered from pain, and
34:09underwent surgeries and skin grafts.
34:12She survived, but was left with scars
34:14and disability. What is particularly
34:18bitter is that the hardware and
34:19software changes planned after the
34:21accidents could have prevented this
34:22event. Nine months later, they had
34:27still not been installed in the machine
34:29working in Yakima. The first known bug
34:32occurred within 8 seconds. The next one
34:34waited until the counter reached 256
34:38and went to zero. Chapter seven, class
34:40one. After another accident, the FDA
34:51decided it had enough. Continuing to
34:54use the Theracs while fixes were still
34:56being worked on by engineers was a
34:58dangerous makeshift solution. On
35:00February 6, 1987, a representative from
35:02the American agency contacted Canadian
35:05authorities. Both sides agreed that the
35:08machines had to be shut down until
35:09permanent safeguards were in place.
35:14Four days later, the FDA demanded that
35:17AECL inform all centers that the Therac
35:19-25 could not operate or be used for
35:21therapy. Exceptions were allowed only
35:27in cases where the need to treat a
35:28specific patient outweighed the risks
35:30of using the machine. And no one was
35:34under the illusion anymore that all the
35:36bugs had been found. In March,
35:38representatives from AECL, all clinics
35:41using the Theracs, and American and
35:43Canadian officials and medical
35:44physicists met at one table. They
35:47analyzed six accidents and prepared a
35:49list of changes. Users demanded, among
35:52other things, an independent software
35:54assessment and a permanent record of
35:56the device’s operation to allow for
35:58the reconstruction of its internal
35:59state after a potential accident. The
36:03AECL representative admitted that tests
36:05of the fixes had been performed, but
36:06they were not documented. He also
36:09stated that an independent software
36:11assessment might prove impossible. The
36:14company had no intention of releasing
36:15the source code. On May 26, the FDA
36:19conditionally approved the fourth
36:21version of the corrective plan and
36:23classified the case as Class I. This
36:25was the highest class of product recall
36:27, used when there is a danger that the
36:29use of a device could cause injury. In
36:33June, a new safety analysis showed that
36:36the plan still did not cover
36:37potentially dangerous areas such as
36:39beam scanning, energy selection, beam
36:41shutdown, and control calibration. A
36:44fifth version of the corrective plan
36:46was needed. On July 21, 1987, AECL
36:49presented its final corrective plan.
36:53From then on, if an error was detected,
36:55the machine would completely abort the
36:56treatment without the possibility of
36:58resuming with the P key. Independent
37:01hardware safeguards were also added to
37:03stop the dangerous beam and prevent it
37:05from firing if the machine was
37:07misconfigured. Enigmatic fault codes
37:09were replaced with understandable
37:11messages, and the function of editing
37:13keys was restricted. The set and reset
37:15commands were changed. Additional
37:17software interlocks were added, and
37:19instructions were improved. The full
37:21implementation of the safeguards, along
37:23with upgrades and safety analysis, took
37:25another year. The final report was
37:27released on November 3, 1988. Two years
37:30and seven months had passed since the
37:32Tyler accident, and three years and
37:34five months since the overdose in
37:36Kennestone. All machines were
37:38re-approved for treating patients.
37:41After the additional safeguards were
37:43introduced, no more accidents involving
37:44the machines were ever recorded. It is
37:47unknown how much longer they remained
37:49in operation. Let’s summarize.
38:01Stories like the Therac-25 fuel
38:03distrust of medicine, and sometimes it
38:05is hard to be surprised by that.
38:08Patients entrusted their lives to
38:09specialists and devices that were
38:11supposed to heal them. Meanwhile, the
38:13manufacturer spent months claiming that
38:15a massive overdose was impossible. It
38:18is easy then to think: if doctors,
38:20engineers, and institutions make
38:22mistakes, perhaps it is safer to look
38:24for help outside the system, from
38:26people offering miracle therapies,
38:28detoxes, healing energy, or vibrations.
38:31However, an unproven method does not
38:33become effective just because modern
38:35medicine can be fallible. The Terak 25
38:38was rebuilt and is back in use for
38:39treatment. Locks were added, programs
38:42were improved, and so on. There have
38:43been no further accidents reported. Of
38:46course, for earlier victims, it's too
38:49late and changes nothing, but let's
38:51remember that the rise in life
38:53expectancy is owed to thousands of such
38:55steps, errors, problems, and later
38:57corrections, clean water, vaccines,
38:59antibiotics, and safe surgeries. You
39:05know, after such accidents, there’s
39:07usually a revised manual, an extra
39:09sensor, a duty to report failures,
39:11millions of minor changes that add up
39:13to extending human life. Medicine must
39:18earn trust by disclosing errors,
39:20investigating them, and turning them
39:21into safeguards that didn't exist
39:23before. This is the only way for people
39:28to trust medicine on a mass scale.
39:31Thank you for the highest proof of
39:33appreciation, which is staying with us
39:34until the end. We are extremely pleased
39:36that there are more and more of you.
39:39Have a good day, stay healthy, and see
39:41you next time.

About this transcript

This page contains the full transcript of Najgroźniejszy błąd w historii radioterapii by OdwróconaLekcja, generated from the public captions YouTube serves with the video. The transcript has 6,105 words across 1,001 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.