Resumen de Computer-aided diagnosis for prostate cancer using multi-parametric mri
PSA to total PSA is computed; if the ratio is higher than 15 %, the case is considered as pathological [206].
A transrectal ultrasound (TRUS) biopsy is carried out for cases which are considered pathological. At least 6 di erent samples are taken randomly from the right and left parts of the 3 di erent prostate zones: apex, median, and base. These samples are further evaluated using the Gleason grading system [90]. The scoring scheme to characterize the biopsy sample is composed of 5 di erent patterns which correspond to grades ranging from 1 to 5. A higher grade is associated with a poorer prognosis [70]. Then, in the Gleason system, 2 scores are assigned corresponding to (i) the grade of the most present tumour pattern, and (ii) the grade of the second most present tumour pattern [70]. A higher Gleason score (GS) indicates a more aggressive tumour [70]. Also, it should be noted that biopsy is an invasive procedure which can result in serious infection or urine retention [46, 100].
Although PSA screening has been shown to improve early detection of CaP [46], its lack of reliability motivates further investigations using MRI-based computer-aided detection and diagnosis (CAD). Two reliable studies|carried out in the United States [12] and in Europe [113, 248] | have attempted to assess the impact of early detection of CaP, with diverging outcomes [46, 104]. The study carried out in Europe1 concluded that PSA screening reduces CaP-related mortality by 21% to 44% [113, 248], while the American2 trial found no such e ect [12]. However, both studies agree that PSA screening su ers from low speci city, with an estimated rate of 36% [247]. Both studies also agree that over-treatment is an issue: decision making regarding treatment is further complicated by diculties in evaluating the aggressiveness and progression of CaP [63].
Hence, new screening methods should be developed with improved speci city of detection as well as more accurate risk assessment (i.e., aggressiveness and progression).
Current research is focused on identifying new biological markers to replace PSAbased screening [25, 28, 188]. Until such research comes to fruition, these needs can be met through active-surveillance strategy using multiparametric MRI (mp-MRI) techniques [107, 187]. An MRI-CAD system, which is an area of active research and forms 1The European randomized study of screening for prostate cancer (ERSSPC) started in the 1990s in order to evaluate the e ect of PSA screening on mortality rate.
2The prostate lung colorectal and ovarian (PLCO) cancer screening trial is carried out in the United States and intends to ascertain the e ects of screening on mortality rate.
4 1.4 CAD systems for CaP the focus of this thesis, can be incorporated into this screening strategy allowing a more systematic and rigorous follow-up.
Another weakness of the current screening strategy lies in the fact that TRUS biopsy does not provide trustworthy results. Due to its \blind" nature, there is a chance of missing aggressive tumours or detecting microfocal \cancers", which in uences the aggressiveness-assessment procedure [193]. As a consequence, over-diagnosis is estimated at up to 30% [95], while missing clinically signi cant CaP is estimated at up 35% [273].
In an e ort to solve both issues, alternative biopsy approaches have been explored.
MRI/ultrasound (US)-guided biopsy has been shown to outperform standard TRUS biopsy [62]. There, mp-MRI images are fused with US images in order to improve localization and aggressiveness assessment to carry out biopsies. Human interaction plays a major role in biopsy sampling which can lead to low repeatability; by reducing potential human errors at this stage, the CAD framework can be used to improve repeatability of examination. CaP detection and diagnosis can bene t from the use of CAD and MRI techniques.
In an e ort to improve the current stage of CaP diagnosis and detection, this thesis is intended to develop the principles of a mp-MRI-CAD system. A description of the di erent MRI modalities is presented in Chap. 2.
