Popular in Formal Sciences

Formal Sciences

Diameter rigidity for Kähler manifolds with positive bisectional curvature

Indian Institute of Science Ved Datar

Formal Sciences

Rank-Based Multivariate Sarmanov for Modeling Dependence

McMaster University Anas Abdallah

Formal Sciences

The Hybrid High-Order Method for Polytopal Meshes

Monash University

Formal Sciences

Financial Informatics: An Information-Based Approach to Asset Pricing

University of Surrey

Formal Sciences

Multiplicative orders of Gauss periods and the arithmetic of real quadratic fields

University of Newcastle

Formal Sciences

Some Numerical Methods for the Hele-Shaw Equations

University of Massachusetts at Amherst

Formal Sciences

Geometric stabilisation via p-adic integration

University of Toronto

Breast Cancer Recurrence Prediction with Machine Learning

View more insights in Sciences - Computer Science - Formal Sciences

View more figures from American University

View the page of Mohammad Owrang

Read the Study

Mohammad Owrang looks at the significance of the non-clinical prognostic factors of age, ethnicity, and marital status in finding the prognosis for breast cancer patients

Image courtesy of interviewee. March 25, 2024

Generating Workspace

Export as PDF

The presentation discusses the importance of medical prognostication, particularly in relation to breast cancer diagnosis and recurrence. The speaker acknowledges the contributions of graduate students to the research and emphasizes the need for accurate predictions regarding complications and survival rates in breast cancer patients.

The speaker outlines key factors affecting breast cancer outcomes, including clinical information (tumor size, grade, lymph node involvement) and non-clinical factors (patient age, race, genetics). The significance of predicting recurrence is highlighted due to breast cancer's high prevalence and associated mortality.

The presentation details the classification of recurrence into local, regional, and distant categories, explaining the various symptoms and implications of each. The research aims to explore new prediction techniques, using data mining and machine learning to identify risk factors and improve prognostic models.

The methodology includes understanding the data attributes, conducting feature selection, and applying several predictive modeling techniques (e.g., decision trees, support vector machines). The final model demonstrates decision trees as the most accurate for predicting breast cancer recurrence, though the speaker notes challenges such as inconsistent data collection across different healthcare facilities and the need for more comprehensive datasets.

Future directions suggest the necessity of standardizing data collection and increasing patient sample sizes to enhance the validity of predictive models in clinical settings. The presentation concludes with a call for ongoing research and collaboration to improve breast cancer treatment and management strategies.

Language:

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.325 --> 00:03.205
I just wanna in advance, I want to thank a couple

00:03.205 --> 00:06.565
of our graduate student, master student who graduated

00:06.625 --> 00:09.645
and then participated in this, paper

00:09.745 --> 00:14.005
and presentation, Farba, Ari and Ginger Shores.

00:14.005 --> 00:15.485
Basically, as far

00:15.485 --> 00:18.965
as the medical prognostication is concerned, it,

00:18.985 --> 00:22.645
it is really very important part of the medicine these days.

00:22.645 --> 00:24.925
Basically, it is, it is the science of

00:25.535 --> 00:28.685
estimating the complication, you know, in, in,

00:28.685 --> 00:31.365
in the disease and the recurrence of the disease,

00:31.745 --> 00:34.645
and be able to make a prediction as you know,

00:34.665 --> 00:37.005
how long the patient can survive, basically.

00:37.065 --> 00:42.045
So it is, I mean, main part of the, treatment plan

00:42.045 --> 00:44.925
because they have to know, the doctors have to know that,

00:44.985 --> 00:48.125
you know, how severe is the breast cancer

00:48.305 --> 00:51.365
for a particular patient and able to make it a plan.

00:51.775 --> 00:55.605
There are several factors affecting, you know, the,

00:55.705 --> 00:57.845
the breast cancer basically, as far

00:57.845 --> 00:59.245
as the diagnosis is concerned.

00:59.745 --> 01:04.685
Uh, clinical information such as tumor size, tumor grade,

01:05.225 --> 01:08.325
number of lymph nodes that are very important.

01:08.675 --> 01:13.245
Some of them are nonclinical, like age of the patient, race

01:13.345 --> 01:15.125
and ethnicity of the patient,

01:15.315 --> 01:18.285
basically sometimes genetic information.

01:18.405 --> 01:19.525
I mean, it has something to do

01:19.525 --> 01:21.325
with the family history and that kind of thing.

01:22.225 --> 01:25.845
So because of the, you know, high prevalence

01:25.985 --> 01:29.805
and mortality of, of breast cancer among women,

01:29.955 --> 01:34.885
basically it is very important to be able to predict

01:35.465 --> 01:38.485
as when, breast cancer can come back

01:38.535 --> 01:39.685
after the treatment.

01:39.745 --> 01:42.125
It, it, it will come back for some people.

01:42.625 --> 01:44.445
So we wanna know, define

01:44.625 --> 01:46.965
or explore new techniques in order

01:47.025 --> 01:48.565
to be able to predict that.

01:48.905 --> 01:51.525
And, and once we have that, then, then people, you know,

01:51.545 --> 01:55.205
the doctors can make a plan to prevent that treatment

01:55.205 --> 01:57.325
that are appropriate in order to,

01:57.325 --> 01:59.445
prevent such thing to recur.

01:59.445 --> 02:02.045
Again, basically the, the main thing, a again,

02:02.105 --> 02:03.605
let me briefly explain that.

02:04.315 --> 02:08.365
Once the breast cancer is diagnosed, then they make a plan

02:08.385 --> 02:10.565
for the treatment based on, you know, the thing

02:10.685 --> 02:13.325
that they discovered that, so they make the treatment,

02:13.325 --> 02:15.845
treatment could be radiation, chemotherapy,

02:16.035 --> 02:18.285
basically hormone therapy,

02:18.465 --> 02:21.325
and in order to remove the cancer cell,

02:22.025 --> 02:23.365
but sometimes some

02:23.365 --> 02:26.405
of them still stay in their skin basically.

02:26.425 --> 02:27.605
And, and then over the years,

02:27.985 --> 02:31.445
or you know, again months, they start to grow

02:31.545 --> 02:33.445
and then, you know, multiply and,

02:33.445 --> 02:35.285
and then suddenly they come back.

02:35.505 --> 02:37.645
That's what their breast cancer recurrence is.

02:38.145 --> 02:39.645
Now when they come back,

02:39.775 --> 02:42.885
there are different ways then they can come back.

02:43.025 --> 02:46.125
If they come back to the same place that they know,

02:46.125 --> 02:48.605
they initial, like, so

02:48.605 --> 02:50.125
where breast cancer was diagnosed,

02:50.585 --> 02:53.245
we call them local breast cancer recurrence.

02:54.265 --> 02:58.365
If they all in the, sort,

02:59.155 --> 03:01.805
near lymph nodes basically, and,

03:01.805 --> 03:04.525
and then they try to infect them basically then they call

03:04.525 --> 03:06.525
them regional, you know, sort of a thing.

03:06.865 --> 03:08.165
Uh, breast cancer recurrence.

03:08.345 --> 03:11.245
And of course in worsts case, if they start to spread

03:11.305 --> 03:14.445
to other part of the body like a lung and kidney

03:14.585 --> 03:17.245
and then bone basically, and,

03:17.245 --> 03:20.405
and then they call it distant, basically distant.

03:20.765 --> 03:23.605
A lot of time they occur for the late breast cancer, like

03:23.605 --> 03:25.285
after 10 years, you know, that kind of thing.

03:25.985 --> 03:28.845
And of course, the doctors have some sort

03:28.845 --> 03:31.445
of a guideline in terms of to recognize which one is which,

03:31.505 --> 03:34.165
you know, so the way, local one, you know, you,

03:34.265 --> 03:36.525
you have, inflammatory breast,

03:36.595 --> 03:38.845
basically redness in there and,

03:38.865 --> 03:42.005
and then sort of where, for the regional, you know, you,

03:42.065 --> 03:45.005
you basically have a lump basically on your neck,

03:45.005 --> 03:47.245
over there or, or sort of way even on the breast,

03:47.245 --> 03:48.845
basically, you can feel it.

03:49.105 --> 03:51.045
The distant one is really very bad.

03:51.065 --> 03:54.125
You know, you have shortness of breaths, basically you,

03:54.185 --> 03:57.605
you lose appetite, you know, basically to,

03:58.265 --> 04:00.325
eat anything and then you lose weight

04:00.425 --> 04:02.525
and then you get, sometimes even in most cases

04:02.525 --> 04:03.685
you get seizure basically.

04:03.785 --> 04:05.925
So, it becomes very critical.

04:06.485 --> 04:10.545
A lot of, things nowadays, people are looking at it,

04:10.545 --> 04:13.945
the early breast cancer, you know, so of recurrence

04:13.945 --> 04:17.425
or late breast cancer recurrence later on, I'll mention that

04:17.525 --> 04:20.105
as part of the, you know, existing, you know, work

04:20.105 --> 04:23.705
that people are doing that there are several work as far

04:23.705 --> 04:25.385
as the breast cancer diagnosis.

04:25.505 --> 04:27.385
A lot of things are about the diagnosis.

04:27.405 --> 04:28.505
We are pretty good with that.

04:28.665 --> 04:31.065
I mean, the existing work, the physician,

04:31.065 --> 04:32.745
they all know how to do that.

04:32.745 --> 04:36.865
Basically, as far as the breast cancer recurrence are,

04:36.965 --> 04:40.105
are concerned, basically there are several world, some

04:40.105 --> 04:42.585
of them, they are looking at a small amount

04:42.605 --> 04:43.905
of the data basically.

04:44.005 --> 04:45.505
And, and then trying,

04:45.505 --> 04:49.345
to model the breast cancer prediction to see that,

04:49.735 --> 04:52.105
what are, you know, who is at risk.

04:52.105 --> 04:56.825
Basically there are existing work that, try

04:56.845 --> 04:59.785
to find what are the factors that

05:00.295 --> 05:02.665
influence the breast cancer recurrence factors,

05:02.805 --> 05:05.585
for instance, ethnicity race or,

05:05.585 --> 05:09.545
or tumor size, tumor grade or, or your eating habit.

05:09.765 --> 05:11.585
You know, lots of other things that, you know,

05:11.585 --> 05:14.665
they wanna see that what really causes, you know,

05:14.665 --> 05:17.225
the breast cancer to come back, basically.

05:17.285 --> 05:20.225
So they are, you know, so we're trying to find factors

05:20.655 --> 05:23.145
that contribute to the breast cancer recurrence.

05:23.715 --> 05:28.665
There are several work that study the data mining

05:28.805 --> 05:32.305
and machine learning techniques to compare them to see that,

05:32.305 --> 05:34.985
like which one is better moderate, basically,

05:35.115 --> 05:37.745
which one can predict, you know,

05:37.925 --> 05:40.705
the breast cancer recurrence basically, you know, and

05:40.705 --> 05:42.785
therefore that would be a good one, you know, for them

05:42.785 --> 05:43.905
to sort of recommend.

05:45.225 --> 05:49.605
Recent years people are looking at two other major thing,

05:49.815 --> 05:51.685
early breast cancer recurrence

05:51.945 --> 05:55.045
and late breast cancer recurrence early.

05:55.155 --> 05:58.085
They are talking about few months, one year,

05:58.265 --> 05:59.565
two years to three years.

05:59.875 --> 06:02.485
Most of them, they, you know, they say that, you know,

06:02.655 --> 06:04.525
after the treatment, all the things,

06:04.905 --> 06:06.245
and then, you know, the,

06:06.305 --> 06:09.045
the breast cancer come back relapse, I mean the,

06:09.235 --> 06:12.805
relapse, you know, survival free basically ends

06:12.905 --> 06:15.765
and then patient is start to show the symptom

06:15.905 --> 06:18.205
for the breast cancer to coming back basically.

06:19.065 --> 06:23.325
And that way basically, you know, they,

06:23.325 --> 06:26.045
they make a plan in terms of what to do, the treatment.

06:26.115 --> 06:28.925
Sometimes this thing happen to younger patient,

06:28.925 --> 06:32.005
basically they could happen that generally, you know,

06:32.295 --> 06:33.805
lymph node, the lymph node

06:33.925 --> 06:36.325
or the protective mechanism if they are infected,

06:36.385 --> 06:37.485
you know, that's what it is.

06:37.505 --> 06:39.725
You know, the breast cancer is a good chance to come back.

06:40.745 --> 06:44.525
The other, things, other, reports

06:44.665 --> 06:47.325
or looking at the late breast cancer, late breast cancer,

06:47.465 --> 06:49.605
we are talking about after 10 years,

06:49.655 --> 06:52.765
after 20 years, I guess for the,

06:52.805 --> 06:54.325
for a long time people always thought

06:54.325 --> 06:55.445
that the physician thought

06:55.445 --> 06:57.885
that breast cancer can come back within five years.

06:58.105 --> 07:00.045
If it doesn't come back within five years, then,

07:00.045 --> 07:02.405
then you are fine, but not is,

07:02.405 --> 07:03.725
this is not the case these days.

07:03.845 --> 07:07.085
I mean, after 20 years even, they have been incidents

07:07.085 --> 07:09.285
that breast cancer can come back and basically,

07:09.545 --> 07:11.845
and those are the one that,

07:11.985 --> 07:14.645
really you are talking about,

07:14.825 --> 07:18.525
the tumor size is very big, like 20 millimeters, that

07:18.525 --> 07:21.205
that's a huge, you know, shoulder weight, size

07:21.305 --> 07:22.445
of the cancer basically.

07:22.625 --> 07:25.045
So those are the ones that a pretty good chance, you know,

07:25.045 --> 07:27.125
after 15 years, 20 years, they come back.

07:27.145 --> 07:30.685
So that, and you can see that it becomes really important

07:30.945 --> 07:33.525
for the, you know, for the people for, you know,

07:34.165 --> 07:38.325
medicine to understand that when this is gonna come back so

07:38.325 --> 07:40.965
that they can hopefully make a proper plan for that

07:41.545 --> 07:44.405
in our study, again, pretty much like, you know, other,

07:44.545 --> 07:47.725
existing work, basically, if you are trying to predict

07:48.385 --> 07:51.165
the breast cancer recurrence, you know,

07:51.195 --> 07:53.405
what are the factors affecting that?

07:53.405 --> 07:58.075
Basically, basically we have

07:58.085 --> 08:02.635
three steps, at least we, we follow the main steps in order

08:02.775 --> 08:04.075
to, you know, accomplish that.

08:04.075 --> 08:06.875
Basically in, in this study, first we wanted

08:06.875 --> 08:08.595
to understand the data.

08:08.875 --> 08:11.835
I mean, it's fundamental, you know, to understand

08:12.545 --> 08:15.235
what are the attributes that we have, what are the features

08:15.235 --> 08:16.435
that we have and,

08:16.435 --> 08:18.955
and how they are related to each other, basically.

08:19.015 --> 08:22.155
So we, we generally, did a little bit

08:22.155 --> 08:24.875
of data visualization, we call that,

08:24.875 --> 08:28.715
which basically means you try to look at it as how these,

08:28.895 --> 08:30.595
how age got something to do

08:30.595 --> 08:33.075
with the breast cancer recurrence, younger age,

08:33.125 --> 08:34.595
older age, middle age.

08:34.625 --> 08:37.795
Basically. You look at that, you look at it the tumor size,

08:37.855 --> 08:40.555
you know how it is related to the breast cancer recurrence,

08:40.895 --> 08:44.115
how the, you know, number of lymph nodes are related to

08:44.115 --> 08:45.155
that basically.

08:45.735 --> 08:49.595
And, and, and then get a better feeling as, you know, what,

08:49.595 --> 08:50.635
what are the data we have?

08:50.655 --> 08:54.115
And of course, it's always in, in, data, medical data.

08:54.125 --> 08:55.725
There are something that are missing

08:55.755 --> 08:57.165
that you have to take care of them.

08:57.165 --> 08:59.885
There are, missing attribute, missing values

08:59.955 --> 09:02.885
that you have to eliminate them because you cannot use them.

09:02.895 --> 09:04.765
There are, you know, missing information there.

09:05.515 --> 09:10.045
Once you have the, the data basically looked at it

09:10.145 --> 09:12.485
and understand a little bit about, you know,

09:12.905 --> 09:14.685
how these things are related to each other,

09:15.235 --> 09:18.045
then we use some techniques called feature

09:18.305 --> 09:19.925
or attribute selection.

09:20.665 --> 09:25.605
Uh, basically it means we wanna look at it among the 32, 35

09:25.745 --> 09:28.645
or four 40, you know, attributes that we have.

09:29.155 --> 09:31.045
What are the one that are more important?

09:31.555 --> 09:33.205
What are the one, I mean, what are the things

09:33.205 --> 09:34.325
that are more important

09:34.545 --> 09:38.485
or lead to having a breast cancer recurrence, basically?

09:39.025 --> 09:41.045
Uh, and, and, and therefore,

09:41.105 --> 09:42.525
we use different techniques.

09:42.585 --> 09:45.845
For instance, recursive, you know, removal

09:45.945 --> 09:47.045
of the, you know, feature.

09:47.225 --> 09:49.725
It means every time we look at it to see

09:49.725 --> 09:52.405
that if something is not significant, we remove them

09:52.625 --> 09:55.165
and then come back again, do one more time, look at it

09:55.165 --> 09:57.605
and say that, okay, you know, based on the data

09:57.635 --> 10:00.445
that we have, another feature is not that important.

10:00.505 --> 10:01.925
We remove that. It could be anything.

10:01.985 --> 10:04.005
It could be age, it could be, you know, sort

10:04.005 --> 10:06.805
of a tumor size, whatever, if they are not that important.

10:07.105 --> 10:12.005
So gradually we wanna have some of them, like six of them,

10:12.105 --> 10:14.565
10 of them, that we think that they are gonna be important.

10:14.865 --> 10:17.725
So that's recursive, you know, technique that we use it.

10:17.975 --> 10:20.405
There is another technique called Pearson, you know,

10:20.425 --> 10:24.925
so the way, a scheme in order to the, so we remove that.

10:24.925 --> 10:29.285
Basically what it does, it looks at it, the features to see

10:29.545 --> 10:31.125
how they are related to each other,

10:31.305 --> 10:32.845
how they are correlating each other.

10:33.105 --> 10:34.605
In other words, two pieces

10:34.705 --> 10:38.485
of the attributes may be really close to each other, and

10:38.485 --> 10:40.805
therefore you don't need both of them.

10:40.945 --> 10:42.245
You can eliminate one of them

10:42.245 --> 10:44.445
because the other one is not going to give you any,

10:44.545 --> 10:45.645
any chance really.

10:45.865 --> 10:48.725
So that's how you sort of reduce some

10:48.725 --> 10:50.205
of the, you know, attributes.

10:50.385 --> 10:52.845
And, and then from that makes it a smaller,

10:52.845 --> 10:55.125
because a smaller set of

10:55.955 --> 10:59.765
good attributes give you a better accuracy when you have too

10:59.765 --> 11:01.725
many, too many attributes there

11:01.725 --> 11:03.285
that participate in breast cancer,

11:03.395 --> 11:04.725
then we call them overfeeding.

11:05.145 --> 11:08.885
And then you really, you don't know which one is, you know,

11:08.915 --> 11:10.445
more important, that kind of thing.

11:10.585 --> 11:13.765
But few of them really become a focused one, and

11:13.765 --> 11:15.445
therefore it gives a better

11:16.005 --> 11:18.165
accuracy when we sort of create our model.

11:18.585 --> 11:21.125
So you do that, there are, again, other technique,

11:21.365 --> 11:23.085
decision tree, you know, that kind of thing.

11:23.405 --> 11:26.045
Sometime they use that in order to sort of,

11:26.555 --> 11:28.725
eliminate some of the irrelevant

11:28.865 --> 11:31.125
or less important, you know, sort of thing.

11:31.675 --> 11:34.525
Even people can look at it.

11:34.605 --> 11:35.845
I mean, scientists can look at it,

11:35.845 --> 11:37.365
the attributes and, and make a decision.

11:37.465 --> 11:41.045
For instance, I look at it patient ID for instance,

11:41.045 --> 11:43.005
patient Id got nothing to do, you know,

11:43.005 --> 11:44.925
with the breast cancer recurrence or,

11:44.985 --> 11:48.085
or most probably the address also is not that important.

11:48.145 --> 11:50.165
You know, it may not, basically,

11:50.525 --> 11:52.605
although you know, where you work could be, you know,

11:52.705 --> 11:56.485
has some significance in there and, and, or,

11:56.545 --> 11:59.325
or different techniques that, for instance, you use in order

11:59.385 --> 12:01.365
to calculate something or,

12:01.385 --> 12:03.445
or to see that what is the size of the,

12:03.465 --> 12:04.685
for instance, the tumor size.

12:04.715 --> 12:06.285
Well, that's not important.

12:06.565 --> 12:10.165
I mean, the, the size of the tumor is important, not

12:10.385 --> 12:11.925
how you get that basically.

12:11.945 --> 12:13.125
So you could eliminate them

12:13.265 --> 12:15.965
and gradually, hopefully you will, you'll be able to,

12:16.105 --> 12:17.645
to remove some of the thing

12:17.665 --> 12:19.085
and have a good set of the things.

12:19.145 --> 12:22.285
So once you have that, we did all these different five,

12:22.385 --> 12:23.445
six different techniques,

12:23.705 --> 12:26.205
and then we took a union of all of them.

12:26.215 --> 12:28.605
Union means combine them to see that

12:28.875 --> 12:32.205
what are the attributes, what are the features that all

12:32.205 --> 12:33.565
of these techniques have in common?

12:33.905 --> 12:37.125
All of them say for instance, you know,

12:37.265 --> 12:39.525
is important, estrogen, that's a receptor.

12:39.955 --> 12:42.605
They all said that the tumor size is important and

12:42.605 --> 12:44.445
therefore we know that, okay, this is important

12:44.475 --> 12:47.325
because it's not like one technique said it is important.

12:47.675 --> 12:50.125
Five of them, they said too, are important.

12:50.275 --> 12:51.805
Then we'll take it as it important.

12:52.235 --> 12:55.685
Therefore, once we have that, we sort of, out

12:55.685 --> 12:59.285
of the 30, 32 attributes, now we have like a five, six

12:59.285 --> 13:03.125
of them that are significantly contributing

13:03.225 --> 13:05.565
to the breast cancer to recur, basically.

13:05.785 --> 13:09.325
So that, that basically, simplify that.

13:09.425 --> 13:11.285
Of course, our, our,

13:11.595 --> 13:13.885
make it our job a little bit easier as far

13:13.885 --> 13:15.845
as prediction model is concerned.

13:16.025 --> 13:17.645
So that's what, for that one,

13:17.945 --> 13:20.605
let me mention one more thing about the,

13:20.865 --> 13:21.885
the prediction thing.

13:22.015 --> 13:23.965
Again, different, model.

13:24.105 --> 13:28.405
We use that support, you know, so do it, vector machine,

13:28.465 --> 13:31.405
we use that decision three, we use that,

13:31.625 --> 13:34.565
and random forest, we use that neural network.

13:34.705 --> 13:37.725
We use that. All of these, basically,

13:37.875 --> 13:41.285
they have different way of looking at the thing.

13:41.945 --> 13:43.125
Our main concept

13:43.505 --> 13:47.805
or concept was that we have a, sort of a data

13:48.075 --> 13:49.765
that was supervised.

13:49.835 --> 13:52.725
Basically we were doing a supervised, you know, sort of a,

13:52.905 --> 13:55.805
machine learning because we knew about the data.

13:55.985 --> 13:58.445
We knew that there was a breast cancer recurrence

13:58.505 --> 14:00.325
and then all these different attributes.

14:00.465 --> 14:03.125
So it was a supervised learning basically.

14:03.465 --> 14:06.005
Uh, and then when we eliminate, you know, some

14:06.005 --> 14:08.805
of these attributes, it makes our training easier

14:08.835 --> 14:11.325
because we have few things, you know,

14:11.345 --> 14:14.165
to teach this machine that you know, how to look at it

14:14.165 --> 14:15.925
and decide the breast cancer recurrence.

14:16.665 --> 14:21.005
Now, after doing all that, our model set

14:21.005 --> 14:22.925
that decision three is the best one

14:23.075 --> 14:24.885
with more accuracy, basically.

14:25.125 --> 14:26.765
I mean, that's the thing. And,

14:26.905 --> 14:29.835
and in fact, one of the problem

14:29.985 --> 14:34.115
that we have is the lack of, as far as the future direction,

14:34.115 --> 14:36.195
conclusion, whatever is concerned, one

14:36.195 --> 14:37.435
of the main problem is

14:37.435 --> 14:40.435
that there is no standard in gathering the data.

14:40.925 --> 14:43.915
Every hospital have their own sort of a set of attributes.

14:44.295 --> 14:46.795
Uh, some people, they have some information, they say, well,

14:46.815 --> 14:49.035
you know, we have information about the ethnicity

14:49.225 --> 14:50.235
race, that kind of thing.

14:50.425 --> 14:52.285
Another dataset does not have that.

14:52.675 --> 14:55.045
Some of them, they use different techniques in order

14:55.065 --> 14:56.285
to gather the information.

14:56.285 --> 14:57.725
Another one does not have that.

14:58.035 --> 15:00.005
Some, most of them, they don't have genetic

15:00.035 --> 15:01.485
information basically.

15:01.585 --> 15:05.845
So in other words, when you don't have a unified

15:06.185 --> 15:09.725
or universal data collection technique,

15:10.155 --> 15:12.885
then it's gonna be hard to define a model that works

15:12.905 --> 15:14.525
for everybody, basically.

15:15.025 --> 15:16.365
So that's, that's the big problem.

15:17.035 --> 15:20.765
Another major issue is the number of patients.

15:21.155 --> 15:24.885
Most of the data that are available out there, they are,

15:25.065 --> 15:28.965
few of them, like there are smaller one three, some

15:28.965 --> 15:31.285
of the existing work they had 1,400.

15:31.635 --> 15:35.085
Ours was not too bad. 2,500, you know, we had that.

15:35.505 --> 15:38.085
But again, there is always missing information, that kind

15:38.085 --> 15:39.125
of things in, in there.

15:39.265 --> 15:42.425
So the thing that we have

15:42.425 --> 15:46.465
to look at it in the future is basically having more data,

15:47.585 --> 15:49.185
a standard in order so

15:49.185 --> 15:53.345
that all everyone use the same standard, to sort the way

15:54.655 --> 15:56.505
getting the data from the patient.

15:57.245 --> 15:59.665
Of course, some of them unfortunately are not available,

15:59.735 --> 16:02.305
like genetic information is not available.

16:02.405 --> 16:04.705
But the literature, of course, you ask a physician,

16:04.735 --> 16:07.665
they all know that if, if a patient has

16:07.695 --> 16:10.985
that means most probably the mother had also breast cancer,

16:11.175 --> 16:13.105
basically because it's a genetic thing.

16:13.365 --> 16:14.985
Uh, it could be there basically,

16:15.325 --> 16:16.945
but that information is not there.

16:17.085 --> 16:19.145
So whatever we decide, if we say

16:19.145 --> 16:20.505
that decision three is the best,

16:20.805 --> 16:23.425
but decision three is the best for our dataset,

16:24.355 --> 16:25.495
not with anybody else.

16:25.675 --> 16:28.015
And, and that's, that's the thing that, you know,

16:28.475 --> 16:29.575
we have to work on it.

16:29.695 --> 16:32.255
I mean, they, you know, sort of way the clinical, you know,

16:32.255 --> 16:34.335
sort of hospital, they have to make sure that, you know,

16:34.335 --> 16:37.375
they adopt some sort of a, universal, you know,

16:37.445 --> 16:40.615
mechanism in order to, you know, create the data so

16:40.615 --> 16:44.015
that we all work on the same, dataset.

16:44.195 --> 16:46.655
And then if we find something

16:46.805 --> 16:48.695
that could be applied in other things,

16:48.995 --> 16:50.175
pretty much the same thing.

16:50.195 --> 16:53.055
For instance, when when you create a vaccine for,

16:53.115 --> 16:55.175
for one place, you know, you wanna see that,

16:55.175 --> 16:58.975
does it work in, in, sort United State and does, or,

16:58.975 --> 17:00.415
or old people or younger people.

17:00.635 --> 17:03.695
That's, that's how it is because it's not a unique thing.

17:03.795 --> 17:05.255
It doesn't work for everybody.

17:05.315 --> 17:07.095
And so you have to have a different thing.

17:07.675 --> 17:12.335
Uh, overall, I guess if I wanna summarize that, it means,

17:12.835 --> 17:15.935
a lot of, I mean, we, we see

17:15.935 --> 17:18.975
that patients are very worried about the, you know,

17:18.975 --> 17:20.335
breast cancer recurrence.

17:20.495 --> 17:22.375
I mean, they, they do the treatment

17:22.435 --> 17:24.255
and then they think that they're all done.

17:24.545 --> 17:26.455
There are some, sort of

17:26.455 --> 17:29.455
with the breast cancer called triple negative breast cancer,

17:30.205 --> 17:33.495
very, you know, sort of a, aggressive sort

17:33.495 --> 17:34.655
of a type of the breast cancer.

17:34.835 --> 17:37.295
And then it sort of makes it very hard as far

17:37.295 --> 17:39.215
as the treatment or concern.

17:39.375 --> 17:41.855
I mean, they, they recur for sure, basically.

17:42.195 --> 17:45.615
Uh, and, and then it is hard really to, treat them,

17:45.895 --> 17:48.455
I guess a little bit of the, you know, again,

17:48.725 --> 17:50.125
What we need more data.

17:50.485 --> 17:51.645
I mean, that's the whole thing.

17:51.705 --> 17:55.485
And, and, you know, not too many of them are available,

17:55.785 --> 17:58.045
you know, outside as a open source basically.

17:58.145 --> 18:01.485
But I mean, this is the things that I'm, I'm sure that,

18:01.485 --> 18:03.245
you know, people are gonna be working on it

18:03.245 --> 18:05.325
because that's, that's what it is, medicine,

18:05.345 --> 18:07.885
and we are looking for every little bit

18:07.885 --> 18:09.605
of these things are gonna help us.

00:00 - I just wanna in advance, I want to thank a couple of our graduate student, master student who graduated and then participated in this, paper and presentation, Farba, Ari and Ginger Shores. Basically, as far as the medical prognostication is concerned, it,

00:18 - it is really very important part of the medicine these days. Basically, it is, it is the science of estimating the complication, you know, in, in, in the disease and the recurrence of the disease, and be able to make a prediction as you know, how long the patient can survive, basically. So it is, I mean, main part of the, treatment plan because they have to know, the doctors have to know that, you know, how severe is the breast cancer for a particular patient and able to make it a plan.

00:51 - There are several factors affecting, you know, the, the breast cancer basically, as far as the diagnosis is concerned. clinical information such as tumor size, tumor grade, number of lymph nodes that are very important. Some of them are nonclinical, like age of the patient, race and ethnicity of the patient, basically sometimes genetic information. I mean, it has something to do with the family history and that kind of thing.

01:22 - So because of the, you know, high prevalence and mortality of, of breast cancer among women, basically it is very important to be able to predict as when, breast cancer can come back after the treatment. It, it, it will come back for some people. So we wanna know, define or explore new techniques in order to be able to predict that. And, and once we have that, then, then people, you know,

01:51 - the doctors can make a plan to prevent that treatment that are appropriate in order to, prevent such thing to recur. Again, basically the, the main thing, a again, let me briefly explain that. Once the breast cancer is diagnosed, then they make a plan for the treatment based on, you know, the thing that they discovered that, so they make the treatment, treatment could be radiation, chemotherapy, basically hormone therapy,

02:18 - and in order to remove the cancer cell, but sometimes some of them still stay in their skin basically. And, and then over the years, or you know, again months, they start to grow and then, you know, multiply and, and then suddenly they come back. That's what their breast cancer recurrence is. Now when they come back, there are different ways then they can come back.

02:43 - If they come back to the same place that they know, they initial, like, so where breast cancer was diagnosed, we call them local breast cancer recurrence. If they all in the, sort, near lymph nodes basically, and, and then they try to infect them basically then they call them regional, you know, sort of a thing. breast cancer recurrence. And of course in worsts case, if they start to spread

03:11 - to other part of the body like a lung and kidney and then bone basically, and, and then they call it distant, basically distant. A lot of time they occur for the late breast cancer, like after 10 years, you know, that kind of thing. And of course, the doctors have some sort of a guideline in terms of to recognize which one is which, you know, so the way, local one, you know, you, you have, inflammatory breast, basically redness in there and,

03:38 - and then sort of where, for the regional, you know, you, you basically have a lump basically on your neck, over there or, or sort of way even on the breast, basically, you can feel it. The distant one is really very bad. You know, you have shortness of breaths, basically you, you lose appetite, you know, basically to, eat anything and then you lose weight and then you get, sometimes even in most cases you get seizure basically.

04:03 - So, it becomes very critical. A lot of, things nowadays, people are looking at it, the early breast cancer, you know, so of recurrence or late breast cancer recurrence later on, I'll mention that as part of the, you know, existing, you know, work that people are doing that there are several work as far as the breast cancer diagnosis. A lot of things are about the diagnosis. We are pretty good with that. I mean, the existing work, the physician,

04:31 - they all know how to do that. Basically, as far as the breast cancer recurrence are, are concerned, basically there are several world, some of them, they are looking at a small amount of the data basically. And, and then trying, to model the breast cancer prediction to see that, what are, you know, who is at risk. Basically there are existing work that, try to find what are the factors that

05:00 - influence the breast cancer recurrence factors, for instance, ethnicity race or, or tumor size, tumor grade or, or your eating habit. You know, lots of other things that, you know, they wanna see that what really causes, you know, the breast cancer to come back, basically. So they are, you know, so we're trying to find factors that contribute to the breast cancer recurrence. There are several work that study the data mining and machine learning techniques to compare them to see that,

05:32 - like which one is better moderate, basically, which one can predict, you know, the breast cancer recurrence basically, you know, and therefore that would be a good one, you know, for them to sort of recommend. Recent years people are looking at two other major thing, early breast cancer recurrence and late breast cancer recurrence early. They are talking about few months, one year, two years to three years.

05:59 - Most of them, they, you know, they say that, you know, after the treatment, all the things, and then, you know, the, the breast cancer come back relapse, I mean the, relapse, you know, survival free basically ends and then patient is start to show the symptom for the breast cancer to coming back basically. And that way basically, you know, they, they make a plan in terms of what to do, the treatment. Sometimes this thing happen to younger patient,

06:28 - basically they could happen that generally, you know, lymph node, the lymph node or the protective mechanism if they are infected, you know, that's what it is. You know, the breast cancer is a good chance to come back. The other, things, other, reports or looking at the late breast cancer, late breast cancer, we are talking about after 10 years, after 20 years, I guess for the, for a long time people always thought

06:54 - that the physician thought that breast cancer can come back within five years. If it doesn't come back within five years, then, then you are fine, but not is, this is not the case these days. I mean, after 20 years even, they have been incidents that breast cancer can come back and basically, and those are the one that, really you are talking about, the tumor size is very big, like 20 millimeters, that

07:18 - that's a huge, you know, shoulder weight, size of the cancer basically. So those are the ones that a pretty good chance, you know, after 15 years, 20 years, they come back. So that, and you can see that it becomes really important for the, you know, for the people for, you know, medicine to understand that when this is gonna come back so that they can hopefully make a proper plan for that in our study, again, pretty much like, you know, other, existing work, basically, if you are trying to predict

07:48 - the breast cancer recurrence, you know, what are the factors affecting that? Basically, basically we have three steps, at least we, we follow the main steps in order to, you know, accomplish that. Basically in, in this study, first we wanted to understand the data. I mean, it's fundamental, you know, to understand what are the attributes that we have, what are the features that we have and,

08:16 - and how they are related to each other, basically. So we, we generally, did a little bit of data visualization, we call that, which basically means you try to look at it as how these, how age got something to do with the breast cancer recurrence, younger age, older age, middle age. Basically. You look at that, you look at it the tumor size, you know how it is related to the breast cancer recurrence, how the, you know, number of lymph nodes are related to

08:44 - that basically. And, and, and then get a better feeling as, you know, what, what are the data we have? And of course, it's always in, in, data, medical data. There are something that are missing that you have to take care of them. There are, missing attribute, missing values that you have to eliminate them because you cannot use them. There are, you know, missing information there. Once you have the, the data basically looked at it

09:10 - and understand a little bit about, you know, how these things are related to each other, then we use some techniques called feature or attribute selection. basically it means we wanna look at it among the 32, 35 or four 40, you know, attributes that we have. What are the one that are more important? What are the one, I mean, what are the things that are more important or lead to having a breast cancer recurrence, basically?

09:39 - and, and, and therefore, we use different techniques. For instance, recursive, you know, removal of the, you know, feature. It means every time we look at it to see that if something is not significant, we remove them and then come back again, do one more time, look at it and say that, okay, you know, based on the data that we have, another feature is not that important. We remove that. It could be anything.

10:01 - It could be age, it could be, you know, sort of a tumor size, whatever, if they are not that important. So gradually we wanna have some of them, like six of them, 10 of them, that we think that they are gonna be important. So that's recursive, you know, technique that we use it. There is another technique called Pearson, you know, so the way, a scheme in order to the, so we remove that. Basically what it does, it looks at it, the features to see how they are related to each other, how they are correlating each other.

10:33 - In other words, two pieces of the attributes may be really close to each other, and therefore you don't need both of them. You can eliminate one of them because the other one is not going to give you any, any chance really. So that's how you sort of reduce some of the, you know, attributes. And, and then from that makes it a smaller, because a smaller set of

10:55 - good attributes give you a better accuracy when you have too many, too many attributes there that participate in breast cancer, then we call them overfeeding. And then you really, you don't know which one is, you know, more important, that kind of thing. But few of them really become a focused one, and therefore it gives a better accuracy when we sort of create our model. So you do that, there are, again, other technique,

11:21 - decision tree, you know, that kind of thing. Sometime they use that in order to sort of, eliminate some of the irrelevant or less important, you know, sort of thing. Even people can look at it. I mean, scientists can look at it, the attributes and, and make a decision. For instance, I look at it patient ID for instance, patient Id got nothing to do, you know, with the breast cancer recurrence or,

11:44 - or most probably the address also is not that important. You know, it may not, basically, although you know, where you work could be, you know, has some significance in there and, and, or, or different techniques that, for instance, you use in order to calculate something or, or to see that what is the size of the, for instance, the tumor size. Well, that's not important. I mean, the, the size of the tumor is important, not

12:10 - how you get that basically. So you could eliminate them and gradually, hopefully you will, you'll be able to, to remove some of the thing and have a good set of the things. So once you have that, we did all these different five, six different techniques, and then we took a union of all of them. Union means combine them to see that what are the attributes, what are the features that all

12:32 - of these techniques have in common? All of them say for instance, you know, is important, estrogen, that's a receptor. They all said that the tumor size is important and therefore we know that, okay, this is important because it's not like one technique said it is important. Five of them, they said too, are important. Then we'll take it as it important. Therefore, once we have that, we sort of, out of the 30, 32 attributes, now we have like a five, six

12:59 - of them that are significantly contributing to the breast cancer to recur, basically. So that, that basically, simplify that. Of course, our, our, make it our job a little bit easier as far as prediction model is concerned. So that's what, for that one, let me mention one more thing about the, the prediction thing. Again, different, model.

13:24 - We use that support, you know, so do it, vector machine, we use that decision three, we use that, and random forest, we use that neural network. We use that. All of these, basically, they have different way of looking at the thing. Our main concept or concept was that we have a, sort of a data that was supervised. Basically we were doing a supervised, you know, sort of a, machine learning because we knew about the data.

13:55 - We knew that there was a breast cancer recurrence and then all these different attributes. So it was a supervised learning basically. and then when we eliminate, you know, some of these attributes, it makes our training easier because we have few things, you know, to teach this machine that you know, how to look at it and decide the breast cancer recurrence. Now, after doing all that, our model set that decision three is the best one

14:23 - with more accuracy, basically. I mean, that's the thing. And, and in fact, one of the problem that we have is the lack of, as far as the future direction, conclusion, whatever is concerned, one of the main problem is that there is no standard in gathering the data. Every hospital have their own sort of a set of attributes. some people, they have some information, they say, well, you know, we have information about the ethnicity

14:49 - race, that kind of thing. Another dataset does not have that. Some of them, they use different techniques in order to gather the information. Another one does not have that. Some, most of them, they don't have genetic information basically. So in other words, when you don't have a unified or universal data collection technique, then it's gonna be hard to define a model that works

15:12 - for everybody, basically. So that's, that's the big problem. Another major issue is the number of patients. Most of the data that are available out there, they are, few of them, like there are smaller one three, some of the existing work they had 1,400. Ours was not too bad. 2,500, you know, we had that. But again, there is always missing information, that kind of things in, in there. So the thing that we have

15:42 - to look at it in the future is basically having more data, a standard in order so that all everyone use the same standard, to sort the way getting the data from the patient. Of course, some of them unfortunately are not available, like genetic information is not available. But the literature, of course, you ask a physician, they all know that if, if a patient has that means most probably the mother had also breast cancer, basically because it's a genetic thing.

16:13 - it could be there basically, but that information is not there. So whatever we decide, if we say that decision three is the best, but decision three is the best for our dataset, not with anybody else. And, and that's, that's the thing that, you know, we have to work on it. I mean, they, you know, sort of way the clinical, you know, sort of hospital, they have to make sure that, you know,

16:34 - they adopt some sort of a, universal, you know, mechanism in order to, you know, create the data so that we all work on the same, dataset. And then if we find something that could be applied in other things, pretty much the same thing. For instance, when when you create a vaccine for, for one place, you know, you wanna see that, does it work in, in, sort United State and does, or, or old people or younger people.

17:00 - That's, that's how it is because it's not a unique thing. It doesn't work for everybody. And so you have to have a different thing. overall, I guess if I wanna summarize that, it means, a lot of, I mean, we, we see that patients are very worried about the, you know, breast cancer recurrence. I mean, they, they do the treatment and then they think that they're all done. There are some, sort of

17:26 - with the breast cancer called triple negative breast cancer, very, you know, sort of a, aggressive sort of a type of the breast cancer. And then it sort of makes it very hard as far as the treatment or concern. I mean, they, they recur for sure, basically. and, and then it is hard really to, treat them, I guess a little bit of the, you know, again, What we need more data. I mean, that's the whole thing.

17:51 - And, and, you know, not too many of them are available, you know, outside as a open source basically. But I mean, this is the things that I'm, I'm sure that, you know, people are gonna be working on it because that's, that's what it is, medicine, and we are looking for every little bit of these things are gonna help us.

Faculti AI can make mistakes