supplemented with computer-aided ultrasonography ... OBJECTIVE. To determine the extent to which computer- ..... Harry Bleiberg is a paid consultant of. AMD.
2008 THE AUTHORS. JOURNAL COMPILATION Urological Oncology
2008 BJU INTERNATIONAL
ACCURACY OF TRUS AND HISTOSCANNING FOR DETECTING SMALL PROSTATE CANCERS BRAECKMAN et al.
BJUI
The accuracy of transrectal ultrasonography supplemented with computer-aided ultrasonography for detecting small prostate cancers
BJU INTERNATIONAL
Johan Braeckman, Philippe Autier*, Cristina Soviany†, Rina Nir†, Dror Nir†, Dirk Michielsen, Karien Treurnicht‡, Michael Jarmulowicz‡, Harry Bleiberg§, Senthil Govindaraju¶ and Mark Emberton¶ UZ Brussel, Vrije Universiteit, §Unit of Epidemiology and Prevention, Screening Clinic, Jules Bordet Institute, Brussels, † Advanced Medical Diagnostics, Waterloo, Belgium, *International Agency for Research on Cancer, Lyon, France, ‡ Bostwick Laboratories, and ¶Comprehensive Biomedical Centre, University College London Hospitals NHS Trust/ University College London, London, UK Accepted for publication 14 May 2008
Study Type – Diagnostic (exploratory cohort) Level of Evidence 2b OBJECTIVE To determine the extent to which computeraided ultrasonography of the prostate (HistoScanningTM, Advanced Medical Diagnostics, Waterloo, Belgium) can identify tumour foci that correspond to a volume of ≥0.50 mL. PATIENTS AND METHODS Between September 2004 and February 2006, 29 men were HistoScanned before scheduled radical prostatectomy. The threedimensional raw (grey-scaled) data required for HistoScanning analysis were acquired by
INTRODUCTION Currently the best imaging method for detecting prostate cancer noninvasively is MRI [1]; whilst current results obtained with MRI are promising [2] and will undoubtedly improve with stronger magnets, the use of endorectal coils and optimization of sequences [3], MRI remains relatively expensive, is not widely available, and obtaining high-quality prostate sequences depends on the expertise and experience of the radiologists and supporting team [4].
1560
transrectal ultrasonography, and analysed using organ-specific tissue-characterization algorithms which form the core of the HistoScanning technology. The HistoScanning analysis results were compared with the histology of the wholemounted prostate, step-sectioned sagittally at 5-mm intervals, and each slide analysed by 5 × 5 mm grid analysis.
lesions were subsequently confirmed to be present but were ≤0.50 mL on histopathological review. Thus, using the clinically accepted volume threshold of 0.50 mL, the sensitivity, specificity, positive and negative predictive value of HistoScanning were 12/12, 13/16 (82%), 12/15 (80%) and 12/12, respectively, for the cancer foci analysed.
RESULTS
CONCLUSIONS
Of 29 patients, 13 had histology unknown to those evaluating the HistoScanning data. With 0.50 mL as the lower threshold for delineating and visualizing cancer volume, HistoScanning correctly predicted the presence of all 12 lesions that were subsequently confirmed to occupy ≥0.50 mL. In addition three lesions were predicted as being present and of ≥0.50 mL. These three
In this preliminary study, HistoScanning accurately detected cancer foci of ≥0.50 mL; these encouraging results will need to be verified in a larger group of patients.
There is therefore a compelling case for developing a technology which uses a platform that is widely available in both hospital and, perhaps more importantly, in ambulatory-care settings where most prostate cancers are diagnosed. If the technology is accurate in identifying prostate cancer lesions at the threshold of 0.50 mL, and at the same time can reliably exclude the presence of clinically significant cancer lesions at or above this size, there is significant scope for its use in improving both the diagnosis and subsequent management of
men who either have prostate cancer or are deemed to be at risk of harbouring it.
KEYWORDS prostate cancer, diagnosis, risk assessment, HistoScanning
HistoScanningTM (Advanced Medical Diagnostics, Waterloo, Belgium) [5] is a technology that detects specific changes in the tissue morphology by extracting and quantifying statistical features from backscattered ultrasonography data. The core of HistoScanning consists of a set of ‘tissuecharacterization algorithms’ described in a previous report [6]. The same report described the ability of prostate HistoScanning to
© JOURNAL COMPILATION
©
2008 THE AUTHORS
2 0 0 8 B J U I N T E R N A T I O N A L | 1 0 2 , 1 5 6 0 – 1 5 6 5 | doi:10.1111/j.1464-410X.2008.07878.x
ACCURACY OF TRUS AND HISTOSCANNING FOR DETECTING SMALL PROSTATE CANCERS
predict the volume of the index cancerous lesion in 14 men with prostate cancer, compared with the estimate of tumour volume using planimetry at the histopathological assessment, where prostate HistoScanning seemed to be accurate in predicting the volume of the index lesion when compared to estimates from histopathology. Although the accuracy of this characterization was deemed to be either equivalent or better than other noninvasive tests such as MRI, the issues of the minimal threshold at which prostate HistoScanning could detect foci of prostate cancer and the accuracy of HistoScanning in predicting volumes of small prostate cancer were not considered. In the present study we address two aspects of prostate HistoScanning, i.e. the extent to which HistoScanning can identify prostate cancer lesions of ≥0.50 mL, and a comparison between the the HistoScanning prediction of individual tumour foci volumes of ≥0.10 mL and total cancer volume (TCV) and the corresponding volumes determined at histopathological review. The identification and prediction of individual prostate cancer lesions with volumes of ≥0.50 mL was used in a recent study testing the accuracy of MRI [2]. TCV, i.e. the sum of the volumes of all the identified cancer lesions, has also been used previously in assessing the ability of biopsies to predict cancer volume [7,8], and in the association between prostate cancer volume and cancer differentiation (i.e. Gleason score) and long-term outcome after treatment (e.g. PSA recurrence) [9,10].
PATIENTS AND METHODS The study protocol was outlined in a previous report [6]; briefly, between September 2004 and February 2006, 29 men diagnosed with prostate cancer thought to be confined to the prostate, and who were clinically attributed to stage T1c and scheduled for radical prostatectomy (RP) at the Academic Hospital, Free Brussels University (Brussels), were HistoScanned before surgery. Approval from the institute’s Ethics Committee was obtained before the study start, and patients agreed to participate in the study by giving their informed consent. Prostate HistoScanning analysis was used in 14 of the 29 patients with no knowledge of the histology results, i.e. the HistoScanning
©
prediction was made while unaware of the histopathological review. One of the 14 patients in the test set had to be excluded because the specimen arising from the RP was damaged and could not be processed to the standards of the reference test. The index test comprised a standardized three-dimensional (3D) examination of the prostate using motorized TRUS in the saggital plane (Hawk, B-K Medical, Copenhagen, Denmark; equipped with an 8658 probe). The data acquired during TRUS were processed by HistoScanning as follows. For a given data file, the 3D matrix representing the TRUS data back-scattered from the prostate volume was segmented. A polygon which followed the prostate contour was drawn manually on the video screen image using the HistoScanning user interface. HistoScanning analysis was then sequentially applied to sub-volumes of this polygonal volume, each sub-volume corresponding to a tissue volume of ≈0.04 mL. An important attribute of the HistoScanning tissue characterization algorithms is that they can be applied sequentially and with no cross-correlation to segmented volumes throughout the prostate. This permits predictions related to the presence or absence of prostate cancer in small and discrete volumes of tissue within the prostate. The volume of cancerous lesions was calculated by summing the sub-volumes present in adjacent locations which were positive by HistoScanning. The distance of the centre of the detected lesion from the rectal wall and from the base of the prostate was measured using known scan parameters and geometry. Its spatial (3D) position within the prostate was established by having a fixed and standardized scan direction (right to left of the patient) with a known fixed angle step (0.2°) between every scanning frame. For the reference test, the processing of the prostate specimen and mounting of slides was done by the pathology department, and comprised sagittal step-sectioning of the whole-mounted RP specimen at 5-mm intervals, and then a 5 × 5 mm grid analysis of these slides using microscopy, following standard processing and haematoxylin-eosin staining. Within each unit square grid, the presence or absence of cancer, the outline of the tumour lesion, and the percentage of the square involved were reported. The volume of malignant lesions identified at histology was determined by Bostwick
Laboratories, London, using Sigma Scan Pro software (Systat Software Inc, San Jose, California, USA). Based on the largest diameter of the lesion, this software calculated the area, and then the volume of each lesion was determined by multiplying this area by the interval of the step-section (typically 5 mm). If a tumour appeared to be associated with several contiguous pathology section blocks then the volumes computed in each individual section were summed to derive the tumour volume. In both the histological and HistoScanning analysis, the location of the centre of each lesion was estimated by reference to the rectal wall and base of the prostate. The corresponding HistoScan volume (a mean of 35 ultrasonogram frames) was linked to each histology block. The volumes differentiated by HistoScanning in each such volume were then compared with the histological findings from the corresponding step-sectioned block, with respect to the size (volume) and location (distance from the rectal wall and base). The matching of the spatial (3D) location of each lesion within the whole prostate was possible as both the TRUS and the step-sectioning had the same starting point (right of the prostate). As both the TRUS and the histological step sectioning were applied in a standardized and methodical manner, the groups of ultrasonogram frames depicting backscattering from each of the sequentially identified 5-mm step-sectioned histology blocks were easily matched. Matching between histology findings and HistoScanning predictions was for lesions of ≥0.10 mL only. After matching the lesions found with both methods, two types of statistical analysis were used. First, the ability of HistoScanning to correctly identify and classify lesions that were determined to be ≥0.50 mL was calculated. Second, for each patient the TCV was computed, i.e. the sum of all individual cancerous volumes of ≥0.10 mL found at histology. The ability of HistoScanning to predict the TCV was examined by summing the volumes of predicted cancerous lesions of ≥0.10 mL found by HistoScanning and comparing both results. The association between volumes was tested using leastsquares linear regression and calculating the Pearson’s product moment coefficient of correlation.
2008 THE AUTHORS
JOURNAL COMPILATION
©
2008 BJU INTERNATIONAL
1561
B R A E C K M A N ET AL.
RESULTS
Figure 1 shows a strong correlation (Pearson r coefficient = 0.99, P < 0.001) between the volumes of the lesions estimated by HistoScanning and of the 28 cancerous lesions of ≥0.10 mL at histology. For each patient, the TCV was computed; Fig. 2 shows that the TCV predicted by HistoScanning was closely correlated with the TCV estimated by histology (Pearson coefficient r = 0.98, P < 0.001). Table 3 shows the detailed results of histology and HistoScanning analysis for each patient. The mean TCV determined by HistoScanning was 10% greater than that from histology; this difference falls within the range expected due to shrinkage of the tumour during fixation of the prostate specimen.
1562
Volume, mL, at histology 0.10–0.19 0.20–0.49 2 1 0 0 1 0 1 0 2 1 0 0 1 2 0 1 0 0 1 1 0 0 0 1 0 1 8 8
≥0.50 2 1 1 1 0 1 0 2 0 2 1 1 0 12
TABLE 1 Number of cancerous foci of ≥0.10 mL found during histological examination of the entire prostate from 13 patients
Total 5 1 2 2 3 1 3 3 0 4 1 2 1 28
Index test, HistoScanning; cancer foci, mL Reference test 0.10–0.49 ≥0.50 Total Histopathology of RP specimen Cancer foci, mL None found 1 0 1 0.10–0.49 12 3 15 ≥0.50 0 12 12 Total 13 15 28
FIG. 1. Prediction of the volume by HistoScanning of 28 cancerous lesions of ≥0.10 mL found at histology in 13 patients. Volume at histology = 0.98 volume by HistoScanning (Pearson r coefficient 0.99, P < 0.001). 10.00
1.00 0.10 0.10
1.00 10.00 Log volume by prostate HistoScanning, cm3
HistoScanning identified correctly those patients (nos 14, 25 and 29) who had a TCV of
