Computer-aided ultrasonography (HistoScanning): a novel technology for locating and characterizing prostate cancer


Johan Braeckman, Academic Hospital, UZ Brussel, Vrije Universiteit Brussel, Laarbeeklaan 101 (1090) Jette, Belgium.



To assess the extent to which prostate HistoScanningTM (PHS), a new ultrasound-based technology that uses computer-aided analysis to quantify tissue disorganization induced by malignant processes, can identify and characterize foci of prostate cancer compared with step-sectioned radical prostatectomy (RP) specimens.


Between September 2004 and February 2006, 29 men had PHS before their scheduled RP. A three-dimensional ultrasound raw-data file was acquired, and PHS analysed regions of interest (ROI) corresponding to tissue volumes of ≈ 0.04 mL. In 13 men the histology was examined on sections of the whole-mount prostate onto which a grid of 5 × 5 mm squares was applied. On a test set of 14 of the 29 patients, PHS analysis was used before knowing the histology results (blinded data), to predict the maximum tumour diameter, focality, laterality and extraprostatic extension (EPE).


Identification and characterization by PHS of the index tumour in the 14 patients in the test set correlated closely (r = 0.95, P < 0.001) with the reference test. The concordance in the attribution of multifocality (present/absent), unilateral/bilateral disease between PHS and histology was 100%. EPE as determined by PHS was attributed to all three pT3a pathological specimens in the blinded paired data. In the same set of data, EPE was attributed to one prostate cancer that on pathological inspection was deemed to be organ-confined (pT2b).


PHS has the potential to identify and characterize prostate cancer foci noninvasively. The precision appears to be sufficient to suggest that PHS might be useful as a triage test for men deemed to be at risk of prostate cancer and who wish to avoid prostate biopsy.


(prostate) HistoScanning


Advanced Medical Diagnostics


region of interest


radical prostatectomy


extraprostatic extension.


The standard method of diagnosing prostate cancer is TRUS-guided biopsy; although there is general consensus on how this procedure should be conducted to maximize prostate cancer detection rates [1], concerns remain about false-negative results on the one hand and poor representation of cancer, both in terms of burden of disease and of grade, on the other [2].

Because of this lack of precision, and of the harms associated with biopsy of the prostate via the rectal mucosa, alternatives have been sought that might safely reassure a man that he does not have clinically significant prostate cancer, or inform him that he is at high risk of harbouring disease. Many of the noninvasive approaches to evaluating the prostate have been promising but to date lack the necessary attributes to challenge TRUS-guided biopsy. Of these technologies, those that have shown the most potential have been associated with image processing [3], elastography [4], MRI [5] and urine analysis [6]. Despite considerable recent advances, none would appear to have the performance characteristics (sensitivity, specificity, positive and negative predictive value) that would make them likely candidates to either replace prostate biopsy or to serve as a triage test of a prostate biopsy for men deemed to be at risk of prostate cancer.

HistoScanningTM (HS, Advanced Medical Diagnostics, AMD, Waterloo, Belgium) is an ultrasound-based technology that has been developed to distinguish cancerous and noncancerous tissues in solid organs. Here we report the findings of the first ‘proof-of-concept’ study that initially adapted the HS technology for distinguishing between cancerous and noncancerous tissues in the prostate, and then tested whether prostate HS (PHS) could effectively detect, locate and estimate the extension of cancerous processes in the prostate.


HS detects specific changes in the tissue morphology by extracting and quantifying statistical features from back-scattered ultrasound data. The core of HS consists of ‘characterization algorithms’ applied on back-scattered ultrasound data (i.e. the ‘radiofrequency, RF, data’ or the ‘raw data’) before they are transformed by the built-in software of ultrasound machines necessary for forming the grey-scale video image. HS works on a personal computer and can be embedded in computers installed in ultrasound machines.

The characterization algorithms exploit the physical changes to sound waves that result from the interaction of the ultrasound beam and the cancer tissue. These can be summarized as energy loss, erratic spatial distribution of energy and increased entropy. One important attribute of these characterization algorithms is that they can be applied in discrete regions of interest (ROIs) throughout the prostate. Thus the presence or absence of prostate cancer can be ascertained within small and discrete volumes of tissue. The identification of cancer within ROIs means that once identified, PHS can spatially orientate the cancer within the gland. This enables both the location of the cancer to be determined as well as its volume.

The present study was divided in two parts; histology data and ultrasound raw data from half of the patients were used by scientists at AMD for calibrating and refining characterization algorithms that could become part of the PHS, i.e. ‘the training set’. During this phase, three characterization algorithms were refined to discriminate back-scattered ultrasound raw data from prostate cancer cells vs data from the background tissue from which cancer arose or spread.

When the PHS prototype was considered ready for use, it was applied by AMD scientists on ultrasound raw data from the second half of the group of patients, with no knowledge of the histology results, i.e. the test set. Fifteen patients were included in the training set and 14 in the test set (Table 1). Initially every fifth and sixth patient was included in the test set, but when the first PHS prototype seemed to be reliable, all remaining patients were included in the test set. Whilst all results are shown in tabular form only those results that were obtained while unaware of the histology were analysed statistically.

Table 1.  The characteristics of the patients who had RP; the 15 patients in the training set are in normal font and the 14 in the test set are in bold
Patient no.Age, yearsSerum PSA level before biopsy (ng/mL)cT stage*Gleason sum biopsy specimenpT stageGleason sum RP specimenInterval, biopsy to RP, days
  • *

    Clinical stage before RP;

  • †Stage after histology examination of the RP specimen;

  • ‡Prostate cancer diagnosed by histology of tissues removed by TURP for BPH. The serum PSA level before biopsy in this patient was in the normal range.

 2669.3cT1c6pT3a6 55
 36012.0cT1c7pT2c6 65
 47011.7cT1c7pT2c6 82
56410.5cT2a 6pT2c7125
6526.6cT2a 7pT2c5 50
 7464.7cT1c6pT2a6 42
 8696.7cT2a8pT3b7 84
 9619.6cT1c6pT2c6 72
10544.4cT1c4pT2c7 91
11665.7cT1c 6pT2c6 40
12445.3cT2a 6pT2c6 42
136420.0cT1c6pT2c5 63
1465  5.70cT1c 6pT2c9122
15647.0cT1c6pT2c5 69
16635.6cT1c6pT2c7 20
17513.7cT1c4pT2a6 90
19586.4cT2c6pT2b6 28
20604.8cT1c5pT3b6 36
2161  1.1cT1a 5pT2c6 67
22559.6cT1c 6pT2b6113
23679.4cT1c 7pT3a7 59
246517.5cT1c 6pT2c6 65
26616.7cT1c 6pT2a7 34
27579.0cT1c 6pT2a6105
286818.0cT1c 6pT3a7 62
29606.8cT1c 6pT3a7 96

Patients were eligible for inclusion in the study if they had histologically confirmed prostate cancer that was thought to be confined to the prostate, clinically attributed a stage of T1c, and who were scheduled for radical prostatectomy (RP). Approval from the Ethics Committee of the Academic Hospital, Free Brussels University, was obtained before starting the study. Between September 2004 and February 2006, 29 patients met the study criteria and agreed to participate in the study by giving their informed consent. During the training phase, potentially eligible patients were not approached consecutively, but according to progress made in the development of the PHS. Only patients 21–29 were included in a short period. The protocol required the patients to be exposed to PHS (the index test) before their scheduled RP. The index test comprised TRUS (Hawk, B-K Medical, Copenhagen, Denmark; equipped with a 8665 3-D probe) and the ultrasound raw data were acquired with the patient lying in the left lateral position. Rotation of the probe was controlled by a motorised unit that rotated at a pre-set rate from the extreme right-hand side to the extreme left-hand side of the gland, and thus raw data were acquired in the sagittal plane. One file containing ultrasound raw data of each prostate was acquired and saved on the hard disk. The data were subsequently transferred to a CD-ROM and sent to AMD for analysis with PHS.

In 13 patients, the PHS results were verified by comparing them with the RP whole-mount histopathology. The specimens were processed according to Montinori et al.[7]. To enable the comparison between PHS analysis and histology, the prostate gland was sectioned in the sagittal plane, into sections of 3–7 mm that were prepared for histology. A grid of 5 × 5 mm squares was applied on each section and histological findings specific of each square were reported on the grids (Fig. 1). The pathology was reported at the level of each of these superimposed grids. This enabled the pathologist to orientate the tumour spatially within each whole mount. As the training phase progressed, detailed histological data became less necessary, and in the remaining 16 patients, the RP was assessed using standard histology.

Figure 1.

Prostate whole-mount RP specimen showing the tissue types in relation to a superimposed grid of 5 × 5 mm squares. The area within each square corresponds to ROIs analysed with HS. The prostate image is reversed (rectal side up) for showing the indications written on the 5 × 5 mm squares. Symbols in squares: GL, Gleason; xC, number of calcifications; IC1, 2, 3, chronic inflammation light, moderate, severe; IA, acute inflammation; ATR, atrophy; VS, seminal vesicle; Pin, prostatic intraepithelial neoplasia; K%, percent of calcification; S, stroma.

PHS uses successive analyses of segmented samples of ultrasound back-scattered raw data representing the volume of the whole gland (Fig. 2). Such raw data is composed of 8-bit grey-level values mapped linearly from the RF data values. This linear mapping from RF value to grey level value required a special transfer function developed by B-K Medical specifically for this study. A ROI was analysed by computing successive two-dimensional matrices of grey level values related to the three-dimensional ROI. The characterization algorithms were successively applied to subunits of such two-dimensional matrixes, each subunit corresponding to a tissue area of ≈ 0.08 cm2, or a tissue volume of ≈ 0.04 mL (the volume approximation assumed that the subunit is located close to the rectal wall, i.e. close to the probe). This visual output is illustrated in Figs 3 and 4.

Figure 2.

Application of the three PHS characterization algorithms on ultrasound raw data corresponding to each 5 × 5 mm square defined by the grids applied on whole-mount prostate slices of patient no. 5, displayed as histograms showing the distribution of numerical patterns related to specific tissue characteristics captured by each algorithm. Comparing one normal and one malignant area in the prostate of this patient resulted in different distributions of numerical patterns, with distributions related to cancerous areas (in red) systematically shifted to the right (higher values) when compared to distributions related to the normal area (in green). Mathematical integration of the distributions provided by the three characterization algorithms allowed the definition of numerical patterns likely to be specific of nonmalignant or of malignant prostatic tissues.

Figure 3.

An example of one sector of the ultrasound volume file from patient no. 9, where the red area corresponds with the malignant scores on the histograms and with the histological analysis.

Figure 4.

An example from one sector of the ultrasound volume file from patient no. 2, showing EPE (white arrows), later confirmed by whole-mount histology.

The following factors were compared between the results from PHS and from the histology of the RP specimen of the 14 men in the test set: (i) The estimated maximum cross-sectional diameter of the predominant (index) tumour as determined by PHS, with the measured maximum cross-sectional diameter on whole-mount histopathology. This assessed the extent to which PHS can discriminate between cancer and non-cancer, and in doing so correctly identify and characterize the largest tumour within the prostate; (ii) the presence or absence of more than one tumour (multifocality), assessing the ability of PHS to reliably exclude other smaller (satellite) lesions when they were not present, as shown on whole-mount pathology; (iii) unilateral vs bilateral disease, assessing the ability of PHS vs whole-mount pathology to exclude prostate cancer on one side of the gland; (iv) the attribution of extraprostatic extension (EPE), assessing whether PHS could discriminate between cancer and non-cancer in tissues other than prostate, e.g. the extraprostatic tissue principally comprises fibrous tissue.


The mean (range) age of the men was 60 (46–70) years; all prostate cancers were deemed impalpable (T1c) by the urologist recruiting the patients into the study (Table 1). The most common Gleason sum attributed at biopsy was 3 + 3 = 6. The mean (range) serum PSA level was 7.7 (3.7–20) ng/mL. There was both up-grading and up-staging when biopsy characteristics were compared with the RP specimens.

Paired data were obtained from both the index (PHS) and reference test (RP specimen; Table 2). Identification and characterization of the index tumour by PHS in patient nos 5, 6, 11, 12, 14, 21–29 (assessed while unaware of the histology) correlated very closely (r = 0.95, P < 0.001) with the maximum cross-sectional diameter of the index tumour, as determined by the pathologist on whole-mount step-sectioning (Fig. 5). Concordance in the attribution of multifocality (present/absent) between the index test and the reference test was 100% in the paired ‘blinded’ data. Concordance in the attribution of unilateral vs bilateral disease between the index and the reference test in blinded samples was also 100%. EPE as determined by PHS was attributed to all three pT3a pathological specimens in the paired blinded data. In patient no. 22, EPE was attributed to the prostate cancer that on pathological inspection was deemed to be organ-confined (pT2b).

Table 2.  The results from PHS and histology for the maximum cross-sectional tumour diameter, multifocality or otherwise, uni- or bilaterality, and pathological stage. The 15 patients in the training set are in normal font and the 14 in the test set in italic
Patient no.Diameter of largest tumour, mmMultifocal atUni- or bilateral atStage at pathologyEPE at PHS*
at pathologyat PHSpathologyPHSpathologyPHS
 110 8nonouniunipT2ano
 7 510.4nonouniunipT2ano
17 710.2nonouniunipT2ano
21 2 3.2yesyesbibipT2cno
25 7 9.2yesyesuniunipT2ano
Figure 5.

The prediction of the diameter of the largest (index) tumour by PHS in the 14 patients of the test set (r = 0.95 (P < 0.0001) for the 13 tumours with a diameter of <30 mm ( r = 0.89, P < 0.0001, for all 14 tumours).


In this pilot study it was apparent that under set conditions PHS could reliably discriminate cancer from non-cancer in men known to have a diagnosis of prostate cancer who subsequently had RP. This discrimination allowed a prediction of essential characteristics of the cancer, using a noninvasive test. The index test appeared to be capable of determining the position of the tumours and their maximum cross-sectional diameter (and therefore their size) with good accuracy. These estimates allowed the status of the patient to be declared for the presence of uni- or multifocality, bilaterality, and provided an apparently reliable estimate of the local staging of the cancer (T2 vs T3).

Before considering both the clinical and scientific implications of the present findings there are important methodological issues that might have introduced bias into the results of this pilot study. The first relates to blinding; most of the PHS outputs in 15 of the 29 patients were obtained after some degree of iterative modification of the algorithms. Whilst these data have been presented for the sake of transparency and completeness, it is likely that the PHS estimates will have ‘over-performed’. The only truly valid comparisons are therefore those in the 14 patients for whom blinding was fully implemented, i.e. the PHS output was derived with no previous knowledge of the pathology report. It is these data alone that have been analysed and on which the conclusions are based.

As this is the first report of PHS it is likely that within the data there are biases related to the accumulation of experience. Case selection, image acquisition and standardization of pathology had small but probably important modifications throughout the study. Whilst these modifications are both acceptable and expected in preliminary work of this type, these were likely to improve the results in the blinded phase of the study.

The reference test differed from the standard pathological processing that most pathologists would recommend for the prostate. Because the PHS images were obtained in the sagittal plane it was necessary to modify the orientation of the RP specimen before slicing. Standard operating procedures usually recommend that the specimen is sliced in the transverse plane from the apex of the prostate to the base. Our pathologists sliced the specimens from left lateral to right lateral, as this was the way that the PHS images were acquired. Whilst a departure from standard processing we do not consider that this would have resulted in any systematic biases being incorporated into the results. It is likely that future studies will acquire the PHS volume file from base of the prostate to the apex, so that this departure from standard reporting can be avoided.

This is the first report of PHS in men with prostate cancer. The results obtained in this small sample are sufficiently promising to suggest that it has the potential to be at least equivalent and possibly better than other noninvasive techniques that were previously used to characterize prostate cancer. Few of these technologies have been assessed against a reference test such as RP specimens. Most have been used to try to increase prostate cancer detection rates by identifying areas of the prostate with a higher than average probability of prostate cancer. Targeted sampling of these areas, in addition to the standard sampling, would, it was hoped, lead to higher detection rates, sampling that was more representative of the cancer, and lower rates of false-negative results.

So far, neither colour-flow mapping and power Doppler ultrasonography, nor contrast-enhanced ultrasonography, have brought convincing improvements [8,9]. The use of elastography is still in a relatively early stage of development, but recent studies support an increase in detection rates on prostate biopsy [4]. Elastography is currently an area of substantial research activity, and there are essential differences between it and PHS. PHS uses a three-dimensional analysis and generates a three-dimensional output; elastography works in two dimensions. The other key difference relates to the requirement of platforms using elastography to involve the operator in identifying abnormal areas (target lesions) for elastographic analysis; PHS requires no such input.

MRI, both with spectroscopy [10,11] and more recently with gadolinium-enhancement, has shown reliable cancer detection rates for lesions of >5 mm in diameter, using RP specimens as the standard [12]. PHS can apparently detect small tumours, and because PHS is a software technology applied to ultrasound data, it is likely to be less expensive and more available than MRI. Further studies will show whether PHS can be as reliable as MRI with gadolinium.

If the present results are replicated in well-controlled multicentre studies, there is the possibility of using this relatively inexpensive, noninvasive test to characterize prostate cancer with reasonable accuracy or, probably more importantly, preclude it with high levels of certainty. The most obvious clinical application would be for evaluating men with a high serum PSA level who are considering a prostate biopsy. If PHS was shown to generate high negative predictive values (somewhere close to of 90% for excluding clinically significant prostate cancer) it would have a useful role in reassuring men deemed at risk of prostate cancer that a biopsy could be avoided. Whether PHS is reliable for identifying relatively small tumours within the prostate, must be studied in conjunction with prostate biopsies to improve detection rates, as the other forms of image processing or enhancement have been evaluated. Other future indications could relate to staging, monitoring of disease, e.g. in regimens of active surveillance, and targeting of therapy in focal or regional treatments.

Future studies will need to address the following research questions. (i) Initially, performance should be assessed in a multicentre setting, to refine the present results. Key outcomes must be used to assess the ability of PHS to detect small tumours and to investigate whether it can discriminate between tumour grades. Subsequent studies will be required to explore its clinical applicability. (ii) In men at risk of prostate cancer, could PHS be used as triage test before a prostate biopsy, to reduce the number of biopsies taken? (iii) In men having a prostate biopsy, could PHS be used to improve cancer detection rates, reduce sampling error, and reduce false-negative biopsy results? (iv) Could PHS replace the role of MRI in local staging of the prostate in men considering radical therapy? (v) Could PHS be used to target focal or regional therapy for men with prostate cancer? (vi) Could PHS be used as a monitoring instrument for men on active surveillance?


We are indebted to Prof Louis Denis (Oncologisch Centrum Antwerpen, Antwerp, Belgium) for his precious guidance and support. We thank B-K Medical (Herlev, Denmark) for providing the equipment and related technical support that allowed the conduct of this study. Mark Emberton is funded by the Comprehensive Biomedical Centre at University College London/University College London Hospital NHS Trust.

Conflict of interest

Dror Nir, Rina Nir and Cristina Soviany are paid developers of the HistoScanning technology on behalf of AMD. Philippe Autier, Johan Braeckman, Lars Egevad, Christian Garbar, Miriam Pipeleers Marichal, and Dirk Michielsen, received neither fee nor stock options from AMD or from B-K Medical.


  1. USA patent application no. 60/574853, provisional application; ‘implementation of ovarian HistoScanning’ filed on 27 May 2004; Pending PCT application HIS-003-PCT/IB2005/001506 ‘Method and device for tissue characterization’; Pending PCT application HIS-002-PCT/IB2004/001127 ‘Method and System for Selecting and Recording Biopsy Sites in a Body Organ’ .