An innovative iterative thresholding algorithm for tumour segmentation and volumetric quantification on SPECT images: Monte Carlo-based methodology and validation

PURPOSE

Positron emission tomography (PET) and single-photon emission computed tomography (SPECT) imaging play an important role in the segmentation of functioning parts of organs or tumours, but an accurate and reproducible delineation is still a challenging task. In this work, an innovative iterative thresholding method for tumour segmentation has been proposed and implemented for a SPECT system. This method, which is based on experimental threshold-volume calibrations, implements also the recovery coefficients (RC) of the imaging system, so it has been called recovering iterative thresholding method (RIThM). The possibility to employ Monte Carlo (MC) simulations for system calibration was also investigated.

METHODS

The RIThM is an iterative algorithm coded using MATLAB: after an initial rough estimate of the volume of interest, the following calculations are repeated: (i) the corresponding source-to-background ratio (SBR) is measured and corrected by means of the RC curve; (ii) the threshold corresponding to the amended SBR value and the volume estimate is then found using threshold-volume data; (iii) new volume estimate is obtained by image thresholding. The process goes on until convergence. The RIThM was implemented for an Infinia Hawkeye 4 (GE Healthcare) SPECT/CT system, using a Jaszczak phantom and several test objects. Two MC codes were tested to simulate the calibration images: SIMIND and SimSet. For validation, test images consisting of hot spheres and some anatomical structures of the Zubal head phantom were simulated with SIMIND code. Additional test objects (flasks and vials) were also imaged experimentally. Finally, the RIThM was applied to evaluate three cases of brain metastases and two cases of high grade gliomas.

RESULTS

Comparing experimental thresholds and those obtained by MC simulations, a maximum difference of about 4% was found, within the errors (+/- 2% and +/- 5%, for volumes > or = 5 ml or < 5 ml, respectively). Also for the RC data, the comparison showed differences (up to 8%) within the assigned error (+/- 6%). ANOVA test demonstrated that the calibration results (in terms of thresholds or RCs at various volumes) obtained by MC simulations were indistinguishable from those obtained experimentally. The accuracy in volume determination for the simulated hot spheres was between -9% and 15% in the range 4-270 ml, whereas for volumes less than 4 ml (in the range 1-3 ml) the difference increased abruptly reaching values greater than 100%. For the Zubal head phantom, errors ranged between 9% and 18%. For the experimental test images, the accuracy level was within +/- 10%, for volumes in the range 20-110 ml. The preliminary test of application on patients evidenced the suitability of the method in a clinical setting.

CONCLUSIONS

The MC-guided delineation of tumor volume may reduce the acquisition time required for the experimental calibration. Analysis of images of several simulated and experimental test objects, Zubal head phantom and clinical cases demonstrated the robustness, suitability, accuracy, and speed of the proposed method. Nevertheless, studies concerning tumors of irregular shape and/or nonuniform distribution of the background activity are still in progress.

Monte Carlo dose calculations using MCNP4C and EGSnrc/BEAMnrc codes to study the energy dependence of the radiochromic film response to beta-emitting sources

The energy dependence of the radiochromic film (RCF) response to beta-emitting sources was studied by dose theoretical calculations, employing the MCNP4C and EGSnrc/BEAMnrc Monte Carlo codes. Irradiations with virtual monochromatic electron sources, electron and photon clinical beams, a (32)P intravascular brachytherapy (IVB) source and other beta-emitting radioisotopes ((188)Re, (90)Y, (90)Sr/(90)Y,(32)P) were simulated. The MD-55-2 and HS radiochromic films (RCFs) were considered, in a planar or cylindrical irradiation geometry, with water or polystyrene as the surrounding medium. For virtual monochromatic sources, a monotonic decrease with energy of the dose absorbed to the film, with respect to that absorbed to the surrounding medium, was evidenced. Considering the IVB (32)P source and the MD-55-2 in a cylindrical geometry, the calibration with a 6 MeV electron beam would yield dose underestimations from 14 to 23%, increasing the source-to-film radial distance from 1 to 6 mm. For the planar beta-emitting sources in water, calibrations with photon or electron clinical beams would yield dose underestimations between 5 and 12%. Calibrating the RCF with (90)Sr/(90)Y, the MD-55-2 would yield dose underestimations between 3 and 5% for (32)P and discrepancies within +/-2% for (188)Re and (90)Y, whereas for the HS the dose underestimation would reach 4% with (188)Re and 6% with (32)P.

Monte Carlo calculation and experimental verification of the photon energy response of tooth enamel in a head-sized plexiglas phantom

The use of electron paramagnetic resonance (EPR) tooth dosimetry for calculation of organ doses requires conversion of the measured absorbed dose in enamel. Before deriving conversion factors from simulation calculations with a realistic anthropomorphic human phantom, in the current study a simplified phantom was chosen to compare EPR measurement and Monte Carlo calculation. The dose response of tooth enamel of molars at various positions inside a cylindrical Plexiglas phantom of head-size was calculated hy Monte Carlo modelling in parallel photon beams of X rays of 63 keV equivalent energy and 60Co gamma rays (1.25 Mev). For X ray exposure, preliminary results of EPR dosimetry with tooth enamel samples prepared from molars irradiated in the phantom were in agreement with calculation. The mean value of the ratio of the measured to the calculated dose was 0.93 +/- 0.08.

Mechanically induced EPR signals in tooth enamel

Sample preparation of tooth enamel for electron paramagnetic resonance (EPR) dosimetry usually involves mechanical operations. The present study shows that mechanical operations performed without water cooling generate a paramagnetic center inducing a stable isotropic EPR signal with g-value of 2.00320 and linewidth of about 0.1 mT. Using EPR spectrum simulation, the similarity between the mechanically induced signal and the signal generated when the enamel is heated in air at a temperature above 600 degrees C was investigated. Results indicate that the mechanically induced signal is related to sample temperature increase during mechanical friction.

Dental radiography: tooth enamel EPR dose assessment from Rando phantom measurements

Electron paramagnetic resonance dosimetry of tooth enamel is now established as a suitable method for individual dose reconstruction following radiation accidents. The accuracy of the method is limited by some confounding factors, among which is the dose received due to medical x-ray irradiation. In the present paper the EPR response of tooth enamel to endoral examination was experimentally evaluated using an anthropomorphic phantom. The dose to enamel for a single exposure of a typical dental examination performed with a new x-ray generation unit working at 65 kVp gave rise to a CO2- signal of intensity similar to that induced by a dose of about 2 mGy of 60Co. EPR measurements were performed on the entire tooth with no attempt to separate buccal and lingual components. Also the dose to enamel for an orthopantomography exam was estimated. It was derived from TLD measurements as equivalent to 0.2 mGy of 60Co. In view of application to risk assessment analysis, in the present work the value for the ratio of the reference dose at the phantom surface measured with TLD to the dose at the tooth measured with EPR was determined.

Comparison of sample preparation and signal evaluation methods for EPR analysis of tooth enamel

In dose reconstruction by EPR dosimetry with teeth various methods are applied to prepare tooth enamel samples and to evaluate the dosimetric signal. A comparison of seven frequently used methods in EPR dosimetry with tooth enamel was performed. The participating Institutes have applied their own procedure to prepare tooth enamel samples and to evaluate the dosimetric signal. The precision of the EPR measurement and the dependence of the estimated dosimetric signal with irradiation up to 1000 mGy were compared. The obtained results are consistent among the different methods. The reproducibility of the dosimetric signal and its estimated relation with the absorbed dose was found to be very close for the applied methods with one possible exception.

Electron arc therapy treatment planning verification with alanine/EPR dosimetry

Alanine/EPR dosimetry was applied to measure dose distributions during electron arc therapy treatments generated by electron beams moving isocentrically over a stated arc. Alanine-polyethylene pellets, prepared at ISS according to the NIST recipe, inserted in a homogeneous cylindrical and anthropomorphic phantoms, were used. Preliminary, alanine response to static electron beams in the (6-20) MeV nominal energy range was studied. Then alanine dosimetry was applied to determine the dose versus the gantry angle at a reference point in a cylindrical homogeneous phantom. Finally, arc therapy treatment planning verification was performed in anthropomorphic phantom.

T2 relaxation of the parotid gland of patients affected by pleomorphic adenoma

The T2 behavior of parotid gland tissue was investigated in 11 patients affected by pleomorphic adenoma. A protocol that was previously set up to define acquisition and post-processing procedures, reaching an accuracy of 2.5% in phantoms and an in vivo long term reproducibility of 0.9-8.5%, was used for the evaluations. The measurements were carried out on a whole body, superconducting imager, using a neck coil as a receiver. Some reference gel samples were imaged together with the patient and used to correct T2 results. The sequence protocol was a multispin-echo, 16 echoes. Signals were fitted with mono and biexponential decay models and an automatic choice of the best model was performed using the two chisquared comparison. Two T2 maps (T2 monoexponential or short T2 component, and long T2 component) and chisquared maps were then produced. Pathologic and normal tissues showed a dominant monoexponential decay with a good level of biexponentiality (16%-27% of total fitted pixels) due to partial volume effects from the liquid content. Concerning the biexponentiality, no significant differences were found between the fitted pixel fraction of normal and pathologic tissue, because the T2 long component of the lesion was related both to the edema and saliva content, but probably the increase in the first compensated the decrease in the second. Chisquared maps showed that most of the lesions presented a monoexponential core surrounded by a biexponential border probably due to a solid component similar to normal tissue with partial volume effects from saliva content. Ninety-five percent confidence intervals for normal tissue were 69.40-87.80 ms (monoexponential relaxation), 38.19-44.67 ms and 285.84-691.28 ms (short and long components of biexponential relaxation). For pathologic tissue they resulted 172.17-275.83 ms, 53.86-89.98 ms and 442.10-814.58 ms. The monoexponential component, mostly present in the core of the lesion, was the parameter that better characterized pathologic tissue. A comparison was performed between normal tissue of patients and normal tissue of volunteers, whose statistics was collected in a previous study with the same evaluation protocol. Results showed no significant differences in the biexponential fitted fractions and the comparison of relaxation times. We conclude that, for tissue characterization, a multiexponential analysis should be carried out in order to improve accuracy and to obtain more reliable results. Moreover, other than relaxation calculations, a topographical analysis of relaxation distribution, using for instance the chisquared maps, might in the future give us more useful information on tissue structure.

Effectiveness of chemical etching for background electron paramagnetic resonance signal reduction in tooth enamel

A method of preparing enamel samples for electron paramagnetic resonance retrospective dosimetry that leads to a reduction of the non-radiation induced signals (background) is proposed. The method is effective in removing residual dentin and in reducing surface paramagnetic centers both native and induced by the sample preparation procedure. The main steps in the method are the use of longwave length ultraviolet light for distinguishing dentin from enamel and of chemical etching with phosphoric acid for removal of surface paramagnetic centers. Several acid concentrations and etching times were tested and the optimization of the method is illustrated in regard to both sample mass loss and background signal amplitude reduction. The best acid concentration was determined to be 42%, leading to a 30% reduction in the background electron paramagnetic resonance signal amplitude per unit mass.