A multiparametric MRI-based CAD system for accurate diagnosis of bladder cancer staging

https://doi.org/10.1016/j.compmedimag.2021.101911Get rights and content

Highlights

  • A multiparametric CAD to differentiate between BC staging (T1 and T2 stages) using T2W- and DW MRIs.

  • A novel automated CNN bladder segmentation framework utilizing an adaptive shape model.

  • Fusion of functional, texture, and geometric features for classification.

  • Features are collected from nested equidistance surfaces (iso-surfaces) from the whole BC volume.

Abstract

Appropriate treatment of bladder cancer (BC) is widely based on accurate and early BC staging. In this paper, a multiparametric computer-aided diagnostic (MP-CAD) system is developed to differentiate between BC staging, especially T1 and T2 stages, using T2-weighted (T2W) magnetic resonance imaging (MRI) and diffusion-weighted (DW) MRI. Our framework starts with the segmentation of the bladder wall (BW) and localization of the whole BC volume (Vt) and its extent inside the wall (Vw). Our segmentation framework is based on a fully connected convolution neural network (CNN) and utilized an adaptive shape model followed by estimating a set of functional, texture, and morphological features. The functional features are derived from the cumulative distribution function (CDF) of the apparent diffusion coefficient. Texture features are radiomic features estimated from T2W-MRI, and morphological features are used to describe the tumors’ geometric. Due to the significant texture difference between the wall and bladder lumen cells, Vt is parcelled into a set of nested equidistance surfaces (i.e., iso-surfaces). Finally, features are estimated for individual iso-surfaces, which are then augmented and used to train and test machine learning (ML) classifier based on neural networks. The system has been evaluated using 42 data sets, and a leave-one-subject-out approach is employed. The overall accuracy, sensitivity, specificity, and area under the receiver operating characteristics (ROC) curve (AUC) are 95.24%, 95.24%, 95.24%, and 0.9864, respectively. The advantage of fusion multiparametric iso-features is highlighted by comparing the diagnostic accuracy of individual MRI modality, which is confirmed by the ROC analysis. Moreover, the accuracy of our pipeline is compared against other statistical ML classifiers (i.e., random forest (RF) and support vector machine (SVM)). Our CAD system is also compared with other techniques (e.g., end-to-end convolution neural networks (i.e., ResNet50).

Introduction

The most recent released statistics of the American Cancer Society showed that urinary bladder cancer (BC) is the fourth most common cancer among men in the US [1]. Early diagnosis of BC helps clinicians select the best treatment method, which depends on four major essential factors. These factors are the BC staging, T, (T1, T2, T3, and T4); BC grading, G, (G1, G2, and G3), lymph node metastasis, N, (N0, N1, N2, and N3), and distant metastasis, M, (M0 and M1) [2]. Depending on the BC staging, the BC treatment can be widely classified into two main types: non-muscle-invasive bladder cancer (NMIBC, stage T1) and muscle-invasive bladder cancer (MIBC, stage T2). By the time, from 20% to 30% of NMIBC advance to MIBC [3]. So early diagnosis is essential to stop BC from advancing. The MIBC rate of aggressiveness and death rate is very high compared to NMIBC [3].

The cystoscopic examination and histological evaluation of bladder sampled tissue, using transurethral resection (TURB), are considered the gold standard for BC detection. Cystoscopy, however, has its own limitations, especially the difficulty of discriminating between malignant lesions and healthy urothelium. Patients with NMIBC are ultimately treated with TURB followed by intravesical chemotherapy [4], [5]. The standard treatment of MIBC patients is radical cystectomy (RC) combined with cisplatin-based neoadjuvant chemotherapy (NAC) [3]. However, RC-NAC has poor prognosis, and metastases develop within two years after RC in about 50% of patients [4]. Therefore, treatment decision, prognosis, and follow-up management of patients with BC depend on the accurate differentiation between NMIBC and MIBC [6], [3], in which this classification relies mainly on BC staging. Cystoscopic examination with pathological evaluation of the resected tissue is the standard reference for differentiation between BC staging between MIBC and NMIBC [5], [7]. From 20% to 30% of BC are incorrectly staged because of variation in performing resection [8], [6]. Using multiparametric modalities and multiple examinations can reduce the diagnostic error by fusing information from different modalities and thus enhance BC diagnosis [9], [10]. However, these methods are time-consuming, invasive, and costly procedures. It is crucial to develop a noninvasive low-cost approach with high accuracy for BC detection to distinguish between MIBC and NMIBC.

In recent years, artificial intelligence (AI) with deep learning (DL) based on medical imaging modalities, such as magnetic resonance imaging (MRI) and computed tomography (CT), have been exploited as essential diagnostic tools for BC detection, tumor staging, and prediction of the treatment response of tumor recurrence [4], [11], [12], [13]. Mainly, MRIs have been found to play a crucial role in making early localization of BC and diagnosis of invasiveness. Radiomics is a method that extracts different sets of features from routine medical imaging, using data characterization algorithms [14], [15]. These radiomic features enable data to be employed in a decision support system to advance clinical diagnostic, prediction performance, and therapeutic response assessment [16], [17]. Various research work on BC diagnosis using MRI modality and radiomic analysis have been conducted using texture features. The latter was extracted from the greyscale pixel values, and high-order derivative maps [18], as well as functional diffusion-weighted MRI (DW-MRI) using apparent diffusion coefficient (ADC) maps [19], [20]. In summary, recent studies demonstrated potential to improve patient care with adequate management, which can reduce the cost of unnecessary investigations and treatment and improve the outcome of therapy [11], [12], [13].

The critical step for determining the BC staging is the tumor-extension position inside the bladder wall (BW). The tumor's location can be localized inside the BW, intravesical, or extending across perivesical to extravesical (see Fig. 1). However, the tumor's size does not affect the BC staging [2]. The big challenge for differentiation between the NMIBC and MIBC is discrimination between T1 and T2 BC stages. As shown in Fig. 1, both of T3 and T4 BC stages have perivesical or extravesical mass from the BW, so it is not difficult to discriminate them from the T1 BC stage. On the other hand, T1 and T2 stages are visually very close, and the significant difference between them is that the T2 BC stage invades the BW muscle, while the T1 BC stage does not [2], [3], [5]. Because of the reasons mentioned above, it is essential to build a computer-aided diagnosis (CAD) system to differentiate between T1 and T2 BC stages using MP-MRI and radiomic analysis. To the best of our knowledge, this is the first study to develop a CAD system to discriminate between T1 and T2 stages instead of NMIBC and MIBC.

There is very little research that has developed a CAD system to differentiate between MIBC and NMIBC [21], [19] using MRI modality, to the best of our knowledge. Those previous research have many limitations. A pathological tumor is segmented manually. The diagnostic accuracy is related to the MIBC and NMIBC classification; the framework did not work in each stage individually. Besides, T3 and T4 BC stages (i.e., MIBC) have perivesical or extravesical mass from the BW, so it is easy to discriminate them from the T1 BC stage (NMIBC). The big challenge is to differentiate between BC stages, especially T1 and T2 BC stages. Additionally, they used in their analysis the whole tumor without giving the importance for any parts. The extracted features do not reflect the physical meaning of the problem using the entire tumor's radiomic characteristics for each modality.

This manuscript's major objective is to propose an automated CAD system for accurate BC staging classification, especially T1 and T2 BC stages (see Fig. 2). The rest of this paper is organized as follows: Section 3 provides details about the study data. The proposed methodology and employed features are fully explained in Section 3, experimental design and results are outlined in Section 4. Finally, discussion and conclusions are given in Sections 5 and 6, respectively.

Section snippets

Bladder MRI data

Forty-two patients who underwent a diagnosis of patients with BC were enrolled in this study after providing consent. The Urology and Nephrology Center, Mansoura University, Egypt, supplied us with the data sets in which the T2-weighted and diffusion MRI were acquired as part of the routine preoperative diagnosis of patients with BC. The patients were enrolled under a protocol approved by the Center's institutional review board (IRB). The T2-weighted and DW-MRI for patients with BC (n = 42,

Methods

In this study, we developed a CAD system (Fig. 2) to discriminate between bladder cancer staging. The overall system applies the following steps to obtain the final diagnosis: Firstly, localization of two VOIs (Vw, Vt) by segmenting the BW with pathology and the BW. After that, extraction of image markers, namely, radiomic and morphological features from the T2-MRI and descriptors of the voxel-wise ADC maps. Finally, diagnosing BC staging utilizes image features to train and test both

Experimental results

The proposed framework was trained and tested on 42 T2W-MRI and DW-MRI subjects with BC, 21 subjects for T1 and T2 BC stages. The MRI scans’ discriminative features were trained and tested using different techniques: neural networks, random forest (RF), and support vector machine (SVM). Besides, to manifest the advantage of our framework, the results for a deep feature (end-to-end fashion) using CNN (i.e., ResNet50) and the approach by Xu et al. [21] were presented. Table 3 summarizes our CAD

Discussion

Early diagnosis of BC helps physicians to select appropriate treatment interventions and thus increase patients’ survival rates. Our goal in this work is the early diagnosis for BC stages, especially T1 and T2 stages. BC staging has many stages, but the most formidable challenge is to differentiate between T1 and T2 BC stages because its visual appearance is almost the same. Also, there is a slight difference in the pathological depth. To the best of our knowledge, this is the first CAD system

Conclusion

In our work, a multiparametric, T2-weighted-magnetic resonance imaging (MRI) and diffusion-weighted MRI, computer-aided diagnostic (MP-CAD) system was developed to discriminate between T1 and T2 BC stages. The MRI scans’ hand-crafted discriminative features were trained and tested for two various techniques: neural networks and statistical ML classifiers, as well as the ResNet50, which was used to validate our proposal results. The diagnostic results approved and confirmed that the proposed

Declaration of interests

None.

Authors’ contribution

K. Hammouda, F. Khalifa, A. Soliman, M. Ghazal, and A. El-Baz: conceptualization, developing the proposed methodology for the analysis, and formal analysis. As well as software, validation, and visualization. K. Hammouda and F. Khalifa: prepared initial draft. M. Ghazal, M. Abou El-Ghar, and A. El-Baz: funding acquisition. M. Abou El-Ghar, and M.A. Badawy: collected local MRI data and participated in formal analysis. K. Hammouda, F. Khalifa, H.E. Darwish, and A. El-Baz: review and edit the

Declaration of Competing Interest

The authors report no declarations of interest.

References (45)

  • J.A. Witjes et al.

    Updated 2016 eau guidelines on muscle-invasive and metastatic bladder cancer

    Eur. Urol.

    (2017)
  • R. Suarez-Ibarrola et al.

    Current and future applications of machine and deep learning in urology: a review of the literature on urolithiasis, renal cell carcinoma, and bladder and prostate cancer

    World J. Urol.

    (2019)
  • S. Wu et al.

    A radiomics nomogram for the preoperative prediction of lymph node metastasis in bladder cancer

    Clin. Cancer Res.

    (2017)
  • M. Takeuchi et al.

    Urinary bladder cancer: diffusion-weighted MR imaging-accuracy for diagnosing T stage and estimating histologic grade

    Radiology

    (2009)
  • P. Turker et al.

    Upstaging of urothelial cancer at the time of radical cystectomy: factors associated with upstaging and its effect on outcome

    BJU Int.

    (2012)
  • S. Kobayashi et al.

    Apparent diffusion coefficient value reflects invasive and proliferative potential of bladder cancer

    J. Magn. Reson. Imaging

    (2014)
  • R. Yang et al.

    Automatic recognition of bladder tumours using deep learning technology and its clinical application

    Int. J. Med. Robot. Comput. Assist. Surg.

    (2020)
  • Lucas, M., Jansen, I., van Leeuwen, T.G., Oddens, J.R., de Bruin, D.M., Marquering, H.A. Deep learning-based recurrence...
  • R.J. Gillies et al.

    Radiomics: images are more than pictures, they are data

    Radiology

    (2016)
  • Y.-Q. Huang et al.

    Development and validation of a radiomics nomogram for preoperative prediction of lymph node metastasis in colorectal cancer

    J. Clin. Oncol.

    (2016)
  • P. Lambin et al.

    Radiomics: the bridge between medical imaging and personalized medicine

    Nat. Rev. Clin. Oncol.

    (2017)
  • X. Xu et al.

    Three-dimensional texture features from intensity and high-order derivative maps for the discrimination between bladder tumors and wall tissues via MRI

    Int. J. Comput. Assist. Radiol. Surg.

    (2017)
  • Cited by (0)

    View full text