TTIs represent the likelihood of cancer at each pixel or voxel within an ROI. To provide the most-rapid processing, we chose to use a LUT to assign values to pixels within the ROI.

Figure 2 shows an SVM-based LUT surface plot for a PSA level of 7.5. The surface plot depicts the relative likelihood of cancer on the vertical axis for the given PSA value. As PSA value increases, all likelihood values increase, but with less emphasis in any particular region of the midband-intercept space, i.e., the LUT surface plot tends to flatten as it rises with increasing PSA values.

The computed spectral parameter values of midband and intercept are entered into the TTI software for each pixel location along with the patient’s PSA level. The combination of midband, intercept, and PSA-level values for each pixel falls on a specific step in the LUT. The value for the relative cancer likelihood at that step in the LUT is returned, and the likelihood value is translated into a color-encoded or gray-scale value for the corresponding pixel in the TTI image. Figure 3 shows an example of a TTI for a prostate that was scanned in situ immediately prior to prostatectomy while the patient was in the operating room. Subsequent prostatectomy histology showed a 12-mm, previously unrecognized, anterior tumor as well as smaller cancer foci and some PIN tissue. The left image in Fig. 3 is the gray-scale TTI; the center image is a midband-parameter image of the same scan plane with superimposed TTI color encoding to display the regions of highest relative likelihoods for cancer; the right image is the approximately corresponding whole-mount histology section, which clearly shows the demarcated anterior tumor and some smaller foci of cancerous and PIN tissue. The planes of the US scans and whole-mount histology are not necessarily identical, but are at the widest gland cross section and clearly overlap in the region of the anterior tumor.

Additional examples of TTIs are the 3D versions of this same prostate, as shown in Fig. 4, and the biopsy-search-window ROI in the TTIs of Fig. 5. In Fig. 4, the most highly suspicious volumes are depicted in red. The representation of the gland in Fig. 4 was generated from a set of parallel scan planes acquired at 5-mm plane separations; 2D TTIs were generated for each plane. The set of 2D TTIs was assembled into a 3D rendering using manual demarcation of the gland surface and TTI color-encoding to demarcate the tumor surfaces automatically. In Fig. 4, the gland is rotated so that it is viewed through the base rather than viewing it more conventionally from the apex as shown in Fig. 3. This 3D rendering could be extremely useful in planning therapy. For example, the rendering clearly shows that a nerve-sparing approach could be applied with a tight margin on the right portion of the gland, but a more generous and cautious one on the left side. The 3D TTI could also provide input for planning focal or differential radiation, cryo-ablation, or HIFU treatments. The illustrative biopsy-guidance TTI windows of Fig. 5 were generated from two separate scans of the same biopsy patient. In this figure five levels of relative cancer likelihood are shown, with green indicating the lowest likelihood, yellow-green a slightly higher likelihood, yellow an intermediate likelihood, orange a moderately high likelihood, and red the highest likelihood. The left image in Fig. 5 was generated from a plane in which the biopsy subsequently proved to be positive; the right image was from a plane that had a negative biopsy. If such images had been available to guide the biopsies in real time, the biopsy needle certainly could have been targeted into the red and orange regions shown in the left image, but the green and yellow-green regions in the right image might have been spared.

Focal treatments for prostate cancer may reduce treatment toxicity without degrading treatment efficacy as long as regions of the gland that harbor cancer are effectively treated. Therefore, a key requirement for focal treatment is reliable imaging of the cancer within the gland. Similarly, biopsies that are guided by images that present the likelihood of cancer can improve the positive yields of core-needle biopsies while simultaneously reducing or eliminating the current need to insert the needle blindly into the gland.

The studies described here use sophisticated, nonlinear methods of classification to distinguish cancerous from noncancerous tissues based on spectrum analysis RF echo signals. At present, these methods rely on ANNs and SVMs, and although more than 600 samples are available from more than 60 patients in the data set cited above, only 17 % of the samples in our most-recent data set are cancerous. Considering the range of possible expressions of cancerous and noncancerous prostatic tissues, further study certainly is required to increase the number of independent samples and to assure that the classifier has sufficient generality. Furthermore, a conservative perspective would consider the patients, and not the biopsy cores, to be the independent samples; to take that perspective into account, a much larger data set is required to fully validate the method and allay concerns regarding the generality of the classifier.

Confidence in reliable ultrasonic TTIs, such as the ones under development and described here, can enable detection and treatment of prostate cancer to advance dramatically. The ultrasonic TTIs illustrated in this chapter can be applied clinically either as pure US techniques using existing digital technology to generate TTIs in real time or they can be used alone or in combination with other modalities (e.g., magnetic resonance imaging) to produce even more powerful methods of imaging prostate cancer. As an example, the mechanical properties sensed by US and exploited in US-based TTIs also can be applied in conjunction with spatially matching chemical properties sensed by magnetic resonance spectroscopy methods, with perfusion and diffusion properties sensed by contrast-enhanced magnetic resonance, or with metabolic properties sensed by positron emission methods. Admittedly, such hybrid methods require reliable means of spatial co-registration, but modern image-processing technology seems to be on the verge of enabling such co-registration to be performed quickly and accurately. The practicing urologist would then have a hybrid TTI based upon real-time US with co-registration of the MRI to produce a fused image that indicates regions of the gland likely to be harboring disease.

The inability of current imaging methods to depict suspicious regions reliably is well recognized and it prevents many patients from choosing the option of active surveillance. A reliable means of imaging cancerous foci would give patients confidence that their diseases can be monitored safely and non-invasively, and consequently, unnecessary treatment of indolent disease could be avoided. Once these QUS-based TTI methods are more fully validated by the ongoing studies cited in the acknowledgments below, clinical implementation should be straightforward.