Mechanical mapping of the prostate in vivo using Dynamic Instrumented Palpation; towards an in vivo strategy for cancer assessment

Abstract

A calibrated palpation sensor has been developed for making instrumented Digital Rectal Examinations (iDREs) with a view to assessing patients for prostate cancer. The instrument measures the dynamic stiffness of the palpable surface of the prostate, and has been trialled on 12 patients in vivo. The patients had been diagnosed with prostate cancer and were scheduled for radical prostatectomy. As far as possible, patients with asymmetric disease were chosen so as to give a variation in gland condition over the palpable surface. The device works by applying an oscillating pressure (force) to a flexible probe whose displacement into the tissue is also measured in order to yield a dynamic stiffness, the static stiffness being incidentally measured at the mean oscillatory force. The device was deployed mounted on the index finger of a urologist and measurements taken at 12-16 positions on each patient using light and firm pressure and palpation frequencies of 1 or 5 Hz. In parallel, conventional DRE assessments were made by a consultant urologist for cancer. After in vivo measurement, the glands were removed and examined histologically with each palpation point being classified as cancerous (C) or not (NC). The work has established the first measurements of static modulus of living prostate tissue to be: 26.8 (13.3) kPa for tissue affected by prostate cancer (C classification), and 24.8 kPa (11.9) for tissue unaffected by cancer (NC classification), values quoted as median (interquartile range). The dynamic properties were characterised by: dynamic modulus, 5.15 kPa (4.86) for the C classification and 4.61 kPa (3.08) for the NC classification and the time lag between force and displacement at 5 Hz palpation frequency, 0.0175 s (0.0078) for the C classification and 0.0186 s (0.0397) for the NC classification, values again quoted as median (interquartile range). With the limited set of features that could be generated, an Artificial Neural Network (ANN) classification yielded a sensitivity of 97%, negative predictive value of 86%, positive predictive value of 67% and accuracy of 70% but with relatively poor specificity (30%). Besides extending the feature set, there are a number of changes in probe design, probing strategy and in mechanics analysis, which are expected to improve the diagnostic capabilities of the method.

Keywords: Prostate cancer diagnosis; dynamic stiffness; elastic modulus; in vivo; relaxation time.

Graphical abstract

Figure 1.

Illustration of the essential principles of DIP: (a) Load extension or stress-strain curve for loading and unloading at a given strain rate, (b) Oscillating force and displacement response at point A on load-displacement curve, (c) Oscillating force and displacement response at point B on load-displacement curve. (a): Loading curve for a given strain rate. (b): Response to an oscillating load of mean 5 units and amplitude 2 units. (c): Response to an oscillating load of mean 2 units and amplitude 1 unit.

Figure 2.

In vivo measurement arrangement. Top left: Probe mounted under clinician’s glove with data acquisition computer in foreground. Top right: Probe elements (photo from Hammer et al.). Bottom: Typical (uncalibrated) load and displacement measurements for a single probe point with actuation frequencies of 5, 10, 15, 20 and 25 Hz.

Figure 3.

Process of classifying in vivo probe points as containing cancer (C) or not (NC). Shown is superior transverse mid-section of excised prostate with approximate positions of the four probe positions on the palpable surface in vivo.

Figure 4.

Example of recorded raw in vivo data at a given point for frequencies of 0.3, 2, 4, 5 and 10 Hz. Top: Force recorded at load cell on actuator output. Bottom: Strain recorded on strain sensitive element on diaphragm. Blue curve: as recorded; Black curves: trend of the average strain over each frequency and resulting de-trended strain signal; Red curves: trend of the average strain over all frequencies and resultingde-trended strain signal.

Figure 5.

ANN confusion matrices for in vivo measurements on patients 11–22, assessed at either 1 or 5 Hz. Class 1 is C and Class 2 is NC and so element 1,1 represents true positives, 1,2, false positives, 2,1 false negatives and 2,2, true negatives.

Figure 6.

ANN confusion matrices for in vivo measurements on patients 15–22, assessed at 5 Hz. Class 1 is C and Class 2 is NC so element 1,1 represents true positives, 1,2, false positives, 2,1 false negatives and 2,2, true negatives.

Figure 7.

Evolution of measured, non-indexed in vivo time lags from Patient 343 over five actuation frequencies compared with ex vivo-based models of Zhang et al. and Barnes et al.