uhh.. no ...

Code:

sample = analogRead(vout); 

blocks on the order of microseconds ... (l00ks like the mega can do 13us if prog correctly default is 100us)

Code:

Serial.println(sample); 

blocks on the order of milliseconds at 9600baud...

 use a scope & trigger a digital i/o to flip hi/lo & you will see your duty cycle ...
    i'll guess .1% dutycycle

Done & with samples .. small partial schematic(just the sensor) & partial code ( droped the sd card & TFT)
http://ralph.ca/?p=707
this camera could also be used (modify the software) to run as a finishing line camera

datasheet for the ILX751A asks for a clk dutycycle of 40-60%  ..
i don't see that happening with that code ..

any luck on slave mode ?

i'm looking to connect a DUE (slave)to a RaspberryPi(master) .. 
and spi has only been implemented on the Pi as master so i figured the Due would be a easier modification ....
 sigh .. time to dig into some data sheets & libs ...

a few comments
the 1Msps is for one channel under very specific conditions
  if you switch channels there is a lag (you will have to look up the details in the data sheet)

IMO best way is to program the PDC with timing generated by the PWM .. with resonable sample code from atmel 
but that will not help you trying to read 80 samples .....

(i'm running at 500,000 samples / sec on one channel, in bursts of ~2k (roughly 4ms/collection ~ pause for 3ms, repeat )

can't comment on the USB , but writing to a SD card "can" be fast  (i'm writing ~500kb/sec),
you need buffer space for sd card lag

these Sony devices have a very simple clocking stucture to follow (as opposed to the National part i was mucking with)

there are really only the 3 steps
pulse low on ROG (once)
pulse on CLK , sample  the data  - repeat 2087 times ....
pause (! found the comment in the data sheet - integreation time is 10ms ! )
and repeat ....
use a scope to check that you have the correct polarities ...

not sure about the MEGA but it appears that you should beable to use the transistor buffer shown in the data sheet so easy to connect ( need clamp diodes for the DUE)
data will look like a stream of 10 bit numbers .... relative to the total amount of light .... 
with junk data at the begining & at the end 

follow up :
i've just had 'limited' success in driving the ILX551A ,
very little change programming change from the other chip (upd3799cy)
& i save myself two pins

no data collection as yet but i can see the output change on the scope
whoo hooo !!

it requires VERY little light (i'm up to 200lines per sec ) & still get resonable responce in a medium dark room on a cloudy day ...
& it appears as the native output is close to max 3.5 volts -- so if i clamp it i may beable to feed the arduino direcly (probably have to buffer it but makes life easier ..... )
have to wait till monday  to get sockets , clamping diodes , & assorted other things ... 

huh?
  simple answer - no

 complex answer -
you write a program & run it on a processor
could you "adc", take the root, "dac" it out  ..yes..

will it work like the circuit ... not likely
will it be a synthisised version of a spice deck .. no

Do you know how use that to get higher frequencies like 25MHz?

not possible
must beable to divide 84MHz by an integer so the limits are
42MHz , 28, 21, 16.8, 14, 12, 10.5 , ....

a fast external converter would be possible - not easy but possible
even if you got the data at full speed can you work with it (see below)

you can skip the early parts of the data - either by tossing it out (sampling & over writing) or delay first sampiling  till the ~first valid bits are available

do you have to run at MAX speed?

other things ( at full speed ) to worry about
 MEGA runs at 16 Mhz (62.5ns) this means if you had to run the sensor at full speed (5Mhz / 200ns) you MIGHT beable to fit 3 risc instructions in the time between samples - i haven't looked at the chips instructions but even if you wrote a super tight loop in asm(NOT C/C++) i'm guessing you would need  5-10 cycles (to read /save data / test if done ) no irqs, no timers

(the above is why i'm using a Due)

and a problem that i'm fussing about - do you know (i don't) when is the exposure time for the chip ? it likely needs 1-100ms to collect data (enough photons to make a image) - do these chips double buffer the data (so you have the 1-100ms to pull the data)  or can you only run one line at a time ...
aaaghhh...

I'd have to use 16 data (+8 for control) should have mentioned TFT not the LCD
have a 7"tft panel on a Due .. (should be plenty of I/Os ..)
& have to do bidirectional 3state level shifts for 16bits
unidirectional level shifts for 8bits
hmmm...
 
 

ok just 2 devices to start

but - has anyone tryed this ?
i'm considering using the 16bits to interface with an PATA/IDE SSD 

200ns not 20ns

no way will you will beable to collect full data at full speed of the sensor
your max conversion is 13μs  (from the data sheet of the Mega)
& likely longer (set up & etc)

you could  decimate & get only 31 samples at full sensor speed (must use the built in sample & hold)
better to slow the sensor to the same speed as the sampling ....

the DUE has a max conversion speed of 1us / sample  & i expect to run at about 1/2 that
(full sensor read taking 2048*2us+overhead ~= 200sensor readings a second )

i have a ILX551A on order (any day now) looks like a 751 without the shutter function
my experiments withthe upd3799cy are at a stand still (though i belive i know the problem & will revisit it ..)
i am using a Due & have some complications that you do not have (3.3v outputs & Adc input )
though i do have more memory to play with ...
i'll publish some code when its up & running ...

other people read threads - not just the posters

"no" is incorrect
sata needs glue  so does pata
if you have the time, inlination & the gpio bits you could use someting like http://www.addonics.com/products/adsaide.php
to connect a sata  drive-  ide/pata is more straight fwd but requires alot more bits

i've connected CF cards directly to a avrMega (not arduinos) (+ 10 yrs ago ..)
 / software is harder than the wireing / but more fat examples exist now than back then ...