def interpFunc(time, deriv=False):
if deriv is False:
basis = nagini.rSplineBasis(time, knots)
interp = np.dot(basis, coefs)
return interp
else:
basis, basisD = nagini.rSplineBasis(time, knots, dot=True)
interp = np.dot(basis, coefs)
deriv = np.dot(basisD, coefs)
return interp, deriv
petData = nagini.reshape4d(pet.get_data()) maskData = mask.get_data().flatten() #Mask pet data petMasked = petData[maskData == 1, :] #Get PET time vetor petTime = info[:, 1] #Caculate whole-brain timecourse petMean = np.mean(petMasked, axis=0) #Get spline knots and basis petKnots = nagini.knotLoc(petTime, args.nKnots[0]) petKnots[0] = 10 petBasis = nagini.rSplineBasis(petTime, petKnots) #Fit spline to whole brain X, _, _, _ = np.linalg.lstsq(petBasis, petMean) #Get spline predictions timePred = np.linspace(0, petTime[-1], 1000) predBasis, predDeriv = nagini.rSplineBasis(timePred, petKnots, dot=True) petPred = np.dot(predBasis, X) petPredD = np.dot(predDeriv, X) #Get timepoints for SUVR window startTime = timePred[np.where(petPredD > 10)[0][0]] endTime = startTime + args.window[0] suvrTime = np.linspace(startTime, endTime, 100)
return interpFunc
#Apply wrapper function
aifFunc = wrapFunc(aifFit[0], aifFit[1], aifFit[2], aifFit[3], aifFit[4],
aifFit[5], aifScale)
else:
#Get basis for AIF spline
if args.nKnots is None:
nKnots = aifC.shape[0]
else:
nKnots = args.nKnots[0]
aifKnots = nagini.knotLoc(aifTime, nKnots, bounds=[5, 95])
aifBasis = nagini.rSplineBasis(aifTime, aifKnots)
#Use scikit learn to do a initial bayesian ridge fit
bModel = linear_model.BayesianRidge(fit_intercept=True)
initFit = bModel.fit(aifBasis[:, 1::], aifC)
initCoefs = np.concatenate((initFit.intercept_[np.newaxis], initFit.coef_))
#Compute absolute studentized residuals
initFitted = np.dot(aifBasis, initCoefs)
initResid = aifC - initFitted
initHat = np.diag(
aifBasis.dot(np.linalg.inv(aifBasis.T.dot(aifBasis)).dot(aifBasis.T)))
initVar = 1.0 / (aifC.shape[0] - aifBasis.shape[1]) * np.sum(
np.power(initResid, 2))
initStudent = np.abs(initResid / np.sqrt(initVar * (1 - initHat)))
else: #Decay correct DCV file to start frame if args.decay == 1: aifC *= np.exp(np.log(2)/122.24*info[0,0]) #Remove decay correction for DCV file else: aifC /= np.exp(np.log(2)/122.24*aifTime) #Get basis for AIF spline if args.nKnots is None: nKnots = aifC.shape[0] else: nKnots = args.nKnots[0] aifKnots = nagini.knotLoc(aifTime,nKnots,bounds=[5,95]) aifBasis = nagini.rSplineBasis(aifTime,aifKnots) #Use scikit learn to do a bayesian style ridge regression for spline fit bModel = linear_model.BayesianRidge(fit_intercept=True) bFit = bModel.fit(aifBasis[:,1::],aifC) aifCoefs = np.concatenate((bFit.intercept_[np.newaxis],bFit.coef_)) aifSpline = np.dot(aifBasis,aifCoefs) #Get interpolation times as start to end of pet scan with aif sampling rate sampTime = np.min(np.diff(aifTime)) interpTime = np.arange(np.floor(startTime[0]),np.ceil(endTime[-1]+sampTime),sampTime) #Make sure we stay within boundaries of pet timing if interpTime[-1] > np.ceil(endTime[-1]): interpTime = interpTime[:-1]
#Get image data petData = nagini.reshape4d(pet.get_data()) maskData = mask.get_data().flatten() #Mask pet data petMasked = petData[maskData==1,:] #Get PET time vetor petTime = info[:,1] #Caculate whole-brain timecourse petMean = np.mean(petMasked,axis=0) #Get spline knots and basis petKnots = nagini.knotLoc(petTime,args.nKnots[0]); petKnots[0] = 10 petBasis = nagini.rSplineBasis(petTime,petKnots) #Fit spline to whole brain X,_,_,_ = np.linalg.lstsq(petBasis,petMean) #Get spline predictions timePred = np.linspace(0,petTime[-1],1000) predBasis,predDeriv = nagini.rSplineBasis(timePred,petKnots,dot=True) petPred = np.dot(predBasis,X); petPredD = np.dot(predDeriv,X) #Get timepoints for SUVR window startTime = timePred[np.where(petPredD>10)[0][0]] endTime = startTime + args.window[0] suvrTime = np.linspace(startTime,endTime,100) #Get new basis
aifC *= np.exp(np.log(2) / 122.24 * info[0, 0]) * np.exp(
np.log(2) / 122.24 * aifTime)
else:
#Decay correct to PET start
aifC *= np.exp(np.log(2) / 122.24 * info[0, 0])
#Set number of knots
if args.nKnots is None:
nKnots = aifTime.shape[0]
else:
nKnots = args.nKnots[0]
#Calculate spline basis
aifKnots = nagini.knotLoc(aifTime, nKnots, bounds=[5, 95])
aifBasis = nagini.rSplineBasis(aifTime, aifKnots)
#Use scikit learn to do a bayesian style ARD regression for fit
bModel = linear_model.ARDRegression(fit_intercept=True)
bFit = bModel.fit(aifBasis[:, 1::], aifC)
aifCoefs = np.concatenate((bFit.intercept_[np.newaxis], bFit.coef_))
#Get basis for interpolating integral of spline at CBV time
pBasis, pBasisI = nagini.rSplineBasis(cbvTime, aifKnots, dDot=True)
#Get integral of input function from start to end of CBV frame
aifInteg = np.dot(pBasisI, aifCoefs)
aifSum = aifInteg[1] - aifInteg[0]
#Get whole-brain pet value
wbPet = np.mean(petMasked)