Question: How does dsolve linearize 1st order ode to 2nd order ode?

Given first order nonlinear ode which is hard to solve using standard methods, the smart dsolve sometimes uses a method where it linearizes the first order ode to a linear second order ode and solves that.

Then with that solution to the linear second order ode, it is able to find the solution of the first order ode (need to resolve the constants of integration to merge them into one, but this part is easy to do).

My question is, how and what method it uses to "linearizes by differentation"? I could not find this in any textbook I have, and not able to see how it does.

Here is an example where it uses this method to solving this first order ode

restart;

Typesetting:-Unsuppress('all'); #always do this.
Typesetting:-Settings(prime=x,'typesetprime'=true); #this says to use y'(x) instead of dy/dx    
Typesetting:-Suppress(y(x)); # this says to use y' and not y'(x)

ode:=diff(y(x),x) = (-y(x)^2+4*a*x)^2/y(x); 
infolevel[dsolve]:=5;
dsolve(ode,y(x), singsol=all);

Tracing says 

So it says, if I understand, that it differentiated the original given first order ode and then linearized the resulting second order ode to the above, which is    y''=-64 a^2 x^2 y - 16 a x y' which is certainly linear second order ode and now can be solved using kovacic algorithm. Now the solution to the first order ode can be obtained.

But when differentiating the first order ode, this is the result

expand(diff(ode,x))

So the question is, did Maple mean it "linearized" the above to y''=-64 a^2 x^2 y - 16 a x y' ? 

If so, then how? Did it use Taylor series? but if so, where it expanded about? Or did it use some other method?  One thing I noticed is that by multiplying the RHS above by y^2 it becomes

And "removing" the nonlinear terms (say expansion is around y=0, so higher order y terms are very small and can be removed) the RHS above becomes

    y'' y^2 = -16 y' a^2 x^2 + 32 a^2 x y

Which is close to Maple shows, but 32 instead of 64, and what about the y^2 on the left side?  Is this method even valid mathematically to do? Since solution that result will be correct only near y=0?

So far, trying to step in the debugger to find how, I was not able to find where it does that but will keep trying.

Any idea what Maple means by linearization to 2nd order and how it does it?

ps. only case I know about, where nonlinear first order ode can be transformed to linear second order ode is the Riccati ode using transformation y=u'/u. But this  first order ode is not Riccati.