- 437
Transcript
Thank you so much. Good afternoon. I'd like thank you for the opportunity, I think to scientific program committee and the Novalis Circle for giving me this opportunity to talk about acoustic neuromas. For my disclosures, I have a consultant role as well as a research grant from Brainlab, as well as a research grant from Stryker, and Baxter for consultants. Andrew Parsa was my mentor and my friend. And he had the saying, he said, "Isaac, don't be afraid of failure. Be afraid of succeeding at the wrong things." And if you would look at the list of my failures, which is very long and very prestigious, I think he'd be quite proud of my list of failures. And the story I hope to tell you today is a story of my failures over the last 15 years about acoustic neuromas.
And it started over a decade ago now. This was published in 2008, looking at hearing preservation in radiosurgery. And this is over a decade, an old saying that hearing preservation was about 60% in radiosurgery. That number was published over 10 years ago and still holds true today. We looked at over 6000 patients being treated with radiosurgery to look at overall hearing preservation, and that for all comers and all kinds of takers and all treatment strategies, that radiosurgery preserved hearing in about 60% of cases. And that now when we consult patients, we tell patients that surgery is a bigger risk to hearing than radiation. But radiation is likely a bigger risk to hearing than doing nothing.
The next place to look that we want look and fail is facial preservation, and that radiosurgery is safer for facial nerve than just for hearing preservation. That over 96%, 97% of the time, that the facial nerve is preserved with radiation, that it's significantly better than doing surgery, but again, that doing nothing is probably safer than doing radiation. And so we also published this. And then this series, this was a cohort of about 2000 patients. But we wanted to ask why. We wanted to fail bigger. So we wanted to ask, why is this hearing preservation different? And so we compare different fractionation strategies for radiosurgery for hearing preservation.
Radiotherapy greater than 5 fractions and this study was principally 28 fractions, or radiosurgery which is less than five fractions. And we looked at the patients that we treated at UCLA, a lot by my colleagues here Dr. Solberg, Dr. Dusalis[SP], Dr. Salch[SP]. And in this series of patients, we compared a typical dose of 12 Gray to 50.4 Gray, which is done 28 fractions. These are the statistics of the comparably compared groups. And what we found was that hearing preservation was about 80% or 90% in radiotherapy in the 28 fractions. And that, 10 years later, when we did this study, again, in a different cohort of patients in single fraction, the hearing preservation was still about 60%. This was very reassuring about the validity and the internal validity of our data.
But that 10 years later at a different institution when we study this same fact that looking for the truth, that hearing preservation was still about 60%. This is from that original paper and this is from the update in 2018. And this was looking at the actuarial curves for that hearing preservation. And we did indeed find that for radiotherapy, hearing preservation was better than single dose radiosurgery, but that curve and that dose effect was only good for about 10 years. And this is a really interesting actuarial data, is that after 10 years that the hearing preservation whether you do radiotherapy or radiosurgery, is probably equivalent. That we saw those hearing preservation curves cross here in the delayed portion of this curve.
So that for the first 10 years, radiotherapy is probably better for hearing preservation than radiosurgery. But that after 10 years, that the hearing preservation after radiation, is probably very similar. And we published these outcomes in neurosurgery here for the outcomes of stereotactic radiosurgery. And so, radiosurgery and radiotherapy have been proven to be very effective treatments for acoustic neuromas, very good at hearing preservation, even better for facial nerve preservation. And our data suggests that radiotherapy may be associated with improved hearing preservation. Well, we wanted to ask why. And so we went back and we looked at all of those contours.
We looked at all of those patients who had hearing preservations and all of their old plans, we pulled them all up, and we did cochlear contouring. We looked at all of their cases. And so how did we do this? Well, the first thing we did was we looked at their MRI and their T2, and their T2 allowed us to find where the cochlea was, and we contour the cochlea. After we did that, we went back and saw, okay, this confirms with the CT scan where the cochlea was. And then we just looked at with a cochlear dose. And we compared that in those two groups. And we found that if the Cochlear dose was lower in single fraction radiosurgery, hearing was better. When the cochlea dose was less than eight Gray in single dose radiosurgery, our hearing preservation was better. But that in multi fraction radiotherapy, the cochlea dose did not alter the outcome.
We kept on looking for ways to fail. My mentor said, fail, just don't succeed at the wrong things. We're looking at the cochlea dose. And so in radiotherapy, it made no difference. But in the single dose radiosurgery, the cochlea dose matters. And this changed our modus operandi. This changed the way I consulted with patients. This changed the way we treated. And so cochlea dose was significantly associated with hearing loss in radiosurgery but not in radiotherapy. And so, for our practice, we tried to keep the cochlea dose less than eight Gray. And that if it could not be constrained to that, we consult the patient about multi fracture. And we also ended up publishing this as well, with the work of Lawrence Chung and Nolan Ang [SP] who put a lot of this data together and so that you can look this data up on your own, but the cochlea dose had a huge impact on hearing preservation.
Well, how does this affect our surgery? This is a publication from Dr. Michael Chugru, [SP] looking at the extent of resection and long term durability of vestibular schwannoma surgery. This is the data from that particular study and this is really the most important slide or the image from that publication in the journal, "Neurosurgery". In the modern era., with radiosurgery, we found that the recurrence rate, the failure rate of surgery and radiation is about 10%. That acoustic neuromas treated radiation controls this tumor about 90% of the time. And this was published about over a decade ago. This is a 15-year story on failure. What's really interesting is that, UCSF most recently with Filty, just published their surgical data, and they also quoted a very similar failure rate for acoustic neuroma surgery of about 9%. That whether you treat acoustic neuromas with radiation, or with surgery, about 9% of them are going to fail.
But somewhere between 9%, 5%, and 10% of these acoustic neuromas, no matter how you treat them, whether you treat them with radiation or surgery, are going to fail, they're going to recur. And that data and that number is staying true, which is really encouraging for validity. For us to know that we may actually know something that's true here. That 90% of these tumors that you can treat with either radiation or surgery, but that 10% of them will come back. Ten percent will come back no matter how you treat them. And this is something that we were touching on very early, about 10 years ago. How does this all work? Like our previous speakers were talking about, if we know that radiation can control tumors in 80% or 90% of surgeries, the argument and the discussion should not be about extent of resection.
We should not be talking about, how much tumor can we take out? It's the wrong question. It's a question that's going to lead to succeeding at the wrong thing. Can you take the whole tumor out? I don't know. You're going to succeed but you're going to succeed at the wrong thing. The real question is, if everyone in this room can agree that radiation or surgery can control the tumor in about 80% or 90% of cases, and if radiation can affect tumor control in 80% or 90% of cases, shouldn't radiation affect 80% to 90% of my acoustic neuroma surgeries? It's a logical question. It's a logical question. It's a question you'll ask if you're looking for failure. If radiation can control 80% to 90% of acoustic neuromas, then radiation should be affecting 80% to 90% of my surgical planning and my acoustic neuroma surgeries.
And so this is how we apply personalized neurosurgery to the actual surgeries. The first place we wanted to do was about eight years ago, I came to Hanover and was inspired to look at the DTI, to see if we could find the facial nerve. And we did this preliminary study in about 40 patients where you can see the acoustic neuroma and you can see the facial nerve around it. You see the yellow nerves curving around the acoustic neuroma. We do this at UCLA, it makes beautiful pictures, really nice pictures. It makes no difference on the clinical outcome, zero difference. That, my outcomes were still the same, my house fragment outcome didn't change. But we published this study, it made really nice pictures.
The second thing is, with this adaptive hybrid surgery, did it make impact on clinical outcomes. Really, or did it just make really nice pictures? And so we performed a study looking at, can we get doctors to idealize what the surgical resection should be and what's the ideal outcome? Because when you tell surgeons, you know what, you don't have to remove the full thing, you should try to preserve facial nerve, just leave a little tumor behind, it's not so simple. It's not so simple. It's like asking someone, how spicy do you like your food? And with neurosurgeons, you can get lots of different answers. And this is what we found in the study. What is the ideal resection? When you ask surgeons just give me the ideal resection, it sounds simple. But your idea of perfection and ideal resection is going to be very, very different than the person sitting right next to you.
What is the ideal resection? And so we performed a study with my patients, surgeries that I had performed. We had done the surgery where if you look in this left corner, you see the initial tumor volume here. And we asked this particular neurosurgeon what does the ideal resection look like? And he said, or she said, "About 24% of the tumor left behind." Over here, this is a second neurosurgeon. This is what they said. And this is the third neurosurgeon, and this is the fourth neurosurgeon. And you get varying responses. This is what the adaptive hybrid surgery software recommended, and this is what I actually left behind. This is the study of my patients, my failures. I resected more tumor than anyone recommended, and I resected a lot more than the ideal resection recommended.
This is what we did. And this is the how that study was performed. This is the initial tumor. This is the recommendation of the software, and these are the four different contours you can see of four different neurosurgeons on this one particular patient. And you can see, there are four different opinions of what the ideal resection should look like, very, very different. No consensus, different pictures. And myself, this is what I contoured. This is what I thought it was. This is what I did. This the failure in my practice, and what I did. And we did this several times with different patients on different studies, and this was the overall result. And you can see that the different neurosurgeons had an idea of what the ideal resection was, always much, much lower than the ideal resection from the computer.
And this is really nice, and this is about as far as I really want to go with study, but I wanted to see how badly I would fail. And so what we really did was we compared this variability to my personal surgeries, and then how they did with their outcomes. How were their house Brackman scores? And we found that if my personal surgery was close to the computer surgery, my facial nerve outcomes was better. I failed. The computer was better than me. I'm looking for ways to fail. We found that every time my surgical resections got closer to the recommended surgical resection, my house Brackman score was better. And whenever I deviate, the further I deviated away from this, the worse my outcome was.
And so we looked at this to look our house Brackman scores, and we published this in "World Neurosurgery" just a few months ago, so that if you'd like, you can look at these details. And we are now moving on to second phase from a mere pilot feasibility study to a much larger patient study, to see if this is generalizable, to see maybe this was also a failure, because we're looking for more and more ways to fail. And this takes me to my last point, which is about the computers and about machine learning. This is a picture that I took off of Twitter and social media, where the Microsoft CEO is here with Stefan Vilsmeier at the Brainlab headquarters, and we're looking at machine learning.
And this is a picture I have from, courtesy of my friend, Chris Ames, looking at predictive analytics and machine learning. And here's what machine learning is all about. There's a really good talk on TED by Kenneth Cukier. He's an economist, an editor at "The Economist". And what they did was they took breast cancer biopsies and they put it into a machine. They didn't give it any rules. No human heuristic, no protocol, no paradigms, just gave it to a machine and said, "Which one of these are breast cancer?" And the computer recognized 12 structural imaging markers, 12 of them, which was fantastic. The problem was that the pathologist and doctors only knew nine of them. The machine could see it better. The machine could see it better.
I, of course, listened to this TED talk and said, "I need to know the source document." So I pulled the source document. I love papers, if you can tell. This is the source document. Go get this paper, it's fantastic. And it's by David Agus. And he's a cancer doctor at USC. And when a UCLA doctor is quoting a USC doctor, you know it has to be true. It's a fantastic paper about machine learning. And it is coming, and it's going to be coming to radiation oncology, and coming to neurosurgery. And this is that first subset is AHS. But this is where it's coming. And this is coming in multiple fields. This is a UCLA doctor now quoting and touting Stanford doctors, where the machine learning is better than the radiologist. And you're like, "This will never happen. This cannot happen in my lifetime."
Well, for those of you in this room who drive cars, and I remember learning to drive a car, when your car starts to spin out and get in trouble, they used to teach me what? You pump the brakes. You pump the brakes. You don't just slam the brakes, you pump the brakes. And now when my daughters learn to drive, they don't tell them to pump the brakes anymore. They say, "Just push as hard as you can." Why? There's a computer in your car. It's called anti-lock brakes. We don't call it a computer, because it's too scary to call it that.. We don't call it auto driving, but you actually already have this. Is that, you have antilock brakes in your car right now that can pump your brakes faster than you can. And that this machine learning is coming. It is coming and going to revolutionize us in our field, radiation oncology and neurosurgery. And it cannot be fought because this is the next generation.
And these young people, these people who did this research that I get to present to you, they're the ones who are leading it, the millennials, the brain tumor department at UCLA, and all the grant funding that supports me. And I'm just going to make a real shameless plug. If you like machine learning, if you like talking about the future of brain tumors, Jeff Weinberg, Monique Oggy [SP] and I are putting the CNS tumor satellite symposium together at the CNS next month in Houston. You should please come. It's going to be a fantastic talk and a great presentation. And if you have any questions, please do not hesitate to Instagram me, Twitter, or Facebook. And if you're old, you can email me or you can text my cell phone. Thank you so much for listening. Appreciate it.
And it started over a decade ago now. This was published in 2008, looking at hearing preservation in radiosurgery. And this is over a decade, an old saying that hearing preservation was about 60% in radiosurgery. That number was published over 10 years ago and still holds true today. We looked at over 6000 patients being treated with radiosurgery to look at overall hearing preservation, and that for all comers and all kinds of takers and all treatment strategies, that radiosurgery preserved hearing in about 60% of cases. And that now when we consult patients, we tell patients that surgery is a bigger risk to hearing than radiation. But radiation is likely a bigger risk to hearing than doing nothing.
The next place to look that we want look and fail is facial preservation, and that radiosurgery is safer for facial nerve than just for hearing preservation. That over 96%, 97% of the time, that the facial nerve is preserved with radiation, that it's significantly better than doing surgery, but again, that doing nothing is probably safer than doing radiation. And so we also published this. And then this series, this was a cohort of about 2000 patients. But we wanted to ask why. We wanted to fail bigger. So we wanted to ask, why is this hearing preservation different? And so we compare different fractionation strategies for radiosurgery for hearing preservation.
Radiotherapy greater than 5 fractions and this study was principally 28 fractions, or radiosurgery which is less than five fractions. And we looked at the patients that we treated at UCLA, a lot by my colleagues here Dr. Solberg, Dr. Dusalis[SP], Dr. Salch[SP]. And in this series of patients, we compared a typical dose of 12 Gray to 50.4 Gray, which is done 28 fractions. These are the statistics of the comparably compared groups. And what we found was that hearing preservation was about 80% or 90% in radiotherapy in the 28 fractions. And that, 10 years later, when we did this study, again, in a different cohort of patients in single fraction, the hearing preservation was still about 60%. This was very reassuring about the validity and the internal validity of our data.
But that 10 years later at a different institution when we study this same fact that looking for the truth, that hearing preservation was still about 60%. This is from that original paper and this is from the update in 2018. And this was looking at the actuarial curves for that hearing preservation. And we did indeed find that for radiotherapy, hearing preservation was better than single dose radiosurgery, but that curve and that dose effect was only good for about 10 years. And this is a really interesting actuarial data, is that after 10 years that the hearing preservation whether you do radiotherapy or radiosurgery, is probably equivalent. That we saw those hearing preservation curves cross here in the delayed portion of this curve.
So that for the first 10 years, radiotherapy is probably better for hearing preservation than radiosurgery. But that after 10 years, that the hearing preservation after radiation, is probably very similar. And we published these outcomes in neurosurgery here for the outcomes of stereotactic radiosurgery. And so, radiosurgery and radiotherapy have been proven to be very effective treatments for acoustic neuromas, very good at hearing preservation, even better for facial nerve preservation. And our data suggests that radiotherapy may be associated with improved hearing preservation. Well, we wanted to ask why. And so we went back and we looked at all of those contours.
We looked at all of those patients who had hearing preservations and all of their old plans, we pulled them all up, and we did cochlear contouring. We looked at all of their cases. And so how did we do this? Well, the first thing we did was we looked at their MRI and their T2, and their T2 allowed us to find where the cochlea was, and we contour the cochlea. After we did that, we went back and saw, okay, this confirms with the CT scan where the cochlea was. And then we just looked at with a cochlear dose. And we compared that in those two groups. And we found that if the Cochlear dose was lower in single fraction radiosurgery, hearing was better. When the cochlea dose was less than eight Gray in single dose radiosurgery, our hearing preservation was better. But that in multi fraction radiotherapy, the cochlea dose did not alter the outcome.
We kept on looking for ways to fail. My mentor said, fail, just don't succeed at the wrong things. We're looking at the cochlea dose. And so in radiotherapy, it made no difference. But in the single dose radiosurgery, the cochlea dose matters. And this changed our modus operandi. This changed the way I consulted with patients. This changed the way we treated. And so cochlea dose was significantly associated with hearing loss in radiosurgery but not in radiotherapy. And so, for our practice, we tried to keep the cochlea dose less than eight Gray. And that if it could not be constrained to that, we consult the patient about multi fracture. And we also ended up publishing this as well, with the work of Lawrence Chung and Nolan Ang [SP] who put a lot of this data together and so that you can look this data up on your own, but the cochlea dose had a huge impact on hearing preservation.
Well, how does this affect our surgery? This is a publication from Dr. Michael Chugru, [SP] looking at the extent of resection and long term durability of vestibular schwannoma surgery. This is the data from that particular study and this is really the most important slide or the image from that publication in the journal, "Neurosurgery". In the modern era., with radiosurgery, we found that the recurrence rate, the failure rate of surgery and radiation is about 10%. That acoustic neuromas treated radiation controls this tumor about 90% of the time. And this was published about over a decade ago. This is a 15-year story on failure. What's really interesting is that, UCSF most recently with Filty, just published their surgical data, and they also quoted a very similar failure rate for acoustic neuroma surgery of about 9%. That whether you treat acoustic neuromas with radiation, or with surgery, about 9% of them are going to fail.
But somewhere between 9%, 5%, and 10% of these acoustic neuromas, no matter how you treat them, whether you treat them with radiation or surgery, are going to fail, they're going to recur. And that data and that number is staying true, which is really encouraging for validity. For us to know that we may actually know something that's true here. That 90% of these tumors that you can treat with either radiation or surgery, but that 10% of them will come back. Ten percent will come back no matter how you treat them. And this is something that we were touching on very early, about 10 years ago. How does this all work? Like our previous speakers were talking about, if we know that radiation can control tumors in 80% or 90% of surgeries, the argument and the discussion should not be about extent of resection.
We should not be talking about, how much tumor can we take out? It's the wrong question. It's a question that's going to lead to succeeding at the wrong thing. Can you take the whole tumor out? I don't know. You're going to succeed but you're going to succeed at the wrong thing. The real question is, if everyone in this room can agree that radiation or surgery can control the tumor in about 80% or 90% of cases, and if radiation can affect tumor control in 80% or 90% of cases, shouldn't radiation affect 80% to 90% of my acoustic neuroma surgeries? It's a logical question. It's a logical question. It's a question you'll ask if you're looking for failure. If radiation can control 80% to 90% of acoustic neuromas, then radiation should be affecting 80% to 90% of my surgical planning and my acoustic neuroma surgeries.
And so this is how we apply personalized neurosurgery to the actual surgeries. The first place we wanted to do was about eight years ago, I came to Hanover and was inspired to look at the DTI, to see if we could find the facial nerve. And we did this preliminary study in about 40 patients where you can see the acoustic neuroma and you can see the facial nerve around it. You see the yellow nerves curving around the acoustic neuroma. We do this at UCLA, it makes beautiful pictures, really nice pictures. It makes no difference on the clinical outcome, zero difference. That, my outcomes were still the same, my house fragment outcome didn't change. But we published this study, it made really nice pictures.
The second thing is, with this adaptive hybrid surgery, did it make impact on clinical outcomes. Really, or did it just make really nice pictures? And so we performed a study looking at, can we get doctors to idealize what the surgical resection should be and what's the ideal outcome? Because when you tell surgeons, you know what, you don't have to remove the full thing, you should try to preserve facial nerve, just leave a little tumor behind, it's not so simple. It's not so simple. It's like asking someone, how spicy do you like your food? And with neurosurgeons, you can get lots of different answers. And this is what we found in the study. What is the ideal resection? When you ask surgeons just give me the ideal resection, it sounds simple. But your idea of perfection and ideal resection is going to be very, very different than the person sitting right next to you.
What is the ideal resection? And so we performed a study with my patients, surgeries that I had performed. We had done the surgery where if you look in this left corner, you see the initial tumor volume here. And we asked this particular neurosurgeon what does the ideal resection look like? And he said, or she said, "About 24% of the tumor left behind." Over here, this is a second neurosurgeon. This is what they said. And this is the third neurosurgeon, and this is the fourth neurosurgeon. And you get varying responses. This is what the adaptive hybrid surgery software recommended, and this is what I actually left behind. This is the study of my patients, my failures. I resected more tumor than anyone recommended, and I resected a lot more than the ideal resection recommended.
This is what we did. And this is the how that study was performed. This is the initial tumor. This is the recommendation of the software, and these are the four different contours you can see of four different neurosurgeons on this one particular patient. And you can see, there are four different opinions of what the ideal resection should look like, very, very different. No consensus, different pictures. And myself, this is what I contoured. This is what I thought it was. This is what I did. This the failure in my practice, and what I did. And we did this several times with different patients on different studies, and this was the overall result. And you can see that the different neurosurgeons had an idea of what the ideal resection was, always much, much lower than the ideal resection from the computer.
And this is really nice, and this is about as far as I really want to go with study, but I wanted to see how badly I would fail. And so what we really did was we compared this variability to my personal surgeries, and then how they did with their outcomes. How were their house Brackman scores? And we found that if my personal surgery was close to the computer surgery, my facial nerve outcomes was better. I failed. The computer was better than me. I'm looking for ways to fail. We found that every time my surgical resections got closer to the recommended surgical resection, my house Brackman score was better. And whenever I deviate, the further I deviated away from this, the worse my outcome was.
And so we looked at this to look our house Brackman scores, and we published this in "World Neurosurgery" just a few months ago, so that if you'd like, you can look at these details. And we are now moving on to second phase from a mere pilot feasibility study to a much larger patient study, to see if this is generalizable, to see maybe this was also a failure, because we're looking for more and more ways to fail. And this takes me to my last point, which is about the computers and about machine learning. This is a picture that I took off of Twitter and social media, where the Microsoft CEO is here with Stefan Vilsmeier at the Brainlab headquarters, and we're looking at machine learning.
And this is a picture I have from, courtesy of my friend, Chris Ames, looking at predictive analytics and machine learning. And here's what machine learning is all about. There's a really good talk on TED by Kenneth Cukier. He's an economist, an editor at "The Economist". And what they did was they took breast cancer biopsies and they put it into a machine. They didn't give it any rules. No human heuristic, no protocol, no paradigms, just gave it to a machine and said, "Which one of these are breast cancer?" And the computer recognized 12 structural imaging markers, 12 of them, which was fantastic. The problem was that the pathologist and doctors only knew nine of them. The machine could see it better. The machine could see it better.
I, of course, listened to this TED talk and said, "I need to know the source document." So I pulled the source document. I love papers, if you can tell. This is the source document. Go get this paper, it's fantastic. And it's by David Agus. And he's a cancer doctor at USC. And when a UCLA doctor is quoting a USC doctor, you know it has to be true. It's a fantastic paper about machine learning. And it is coming, and it's going to be coming to radiation oncology, and coming to neurosurgery. And this is that first subset is AHS. But this is where it's coming. And this is coming in multiple fields. This is a UCLA doctor now quoting and touting Stanford doctors, where the machine learning is better than the radiologist. And you're like, "This will never happen. This cannot happen in my lifetime."
Well, for those of you in this room who drive cars, and I remember learning to drive a car, when your car starts to spin out and get in trouble, they used to teach me what? You pump the brakes. You pump the brakes. You don't just slam the brakes, you pump the brakes. And now when my daughters learn to drive, they don't tell them to pump the brakes anymore. They say, "Just push as hard as you can." Why? There's a computer in your car. It's called anti-lock brakes. We don't call it a computer, because it's too scary to call it that.. We don't call it auto driving, but you actually already have this. Is that, you have antilock brakes in your car right now that can pump your brakes faster than you can. And that this machine learning is coming. It is coming and going to revolutionize us in our field, radiation oncology and neurosurgery. And it cannot be fought because this is the next generation.
And these young people, these people who did this research that I get to present to you, they're the ones who are leading it, the millennials, the brain tumor department at UCLA, and all the grant funding that supports me. And I'm just going to make a real shameless plug. If you like machine learning, if you like talking about the future of brain tumors, Jeff Weinberg, Monique Oggy [SP] and I are putting the CNS tumor satellite symposium together at the CNS next month in Houston. You should please come. It's going to be a fantastic talk and a great presentation. And if you have any questions, please do not hesitate to Instagram me, Twitter, or Facebook. And if you're old, you can email me or you can text my cell phone. Thank you so much for listening. Appreciate it.