Dr. Rakesh Shiradkar Profile

Dr. Rakesh Shiradkar

at Emory Univ

SPIE Involvement:

Author

Proceedings Article | 4 April 2022 Presentation + Paper

Evaluating the sensitivity of deep learning to inter-reader variations in lesion delineations on bi-parametric MRI in identifying clinically significant prostate cancer

Ansh Roge, Amogh Hiremath, Michael Sobota, Sree Harsha Tirumani, Leonardo Kayat Bittencourt, Justin Ream, Ryan Ward, Halimat Olaniyan, Sadhna Verma, Andrei Purysko, Anant Madabhushi, Rakesh Shiradkar

Proceedings Volume 12033, 120330Z (2022) https://doi.org/10.1117/12.2613245

KEYWORDS: Principal component analysis, Magnetic resonance imaging, Prostate cancer, Biopsy, Medical research, Pathology, Prostate, Performance modeling, Diffusion, Convolutional neural networks

Read Abstract +

Deep learning based convolutional neural networks (CNNs) for prostate cancer (PCa) risk stratification employ radiologist delineated regions of interest (ROIs) on MRI. These ROIs contain the reader’s interpretation of the region of PCa. Variations in reader annotations change the features that are extracted from the ROIs, which may in turn affect classification performance of CNNs. In this study, we sought to analyze the effect of variations in inter-reader delineations of PCa ROIs on training of CNNs with regards to distinguishing clinically significant (csPCa) and insignificant PCa (ciPCa). We employed 180 patient studies (n=274 lesions) from 3 cohorts who underwent 3T multi-parametric MRI followed by MRI-targeted biopsy and/or radical prostatectomy. ISUP Gleason grade groups (GGG) obtained from pathology were used to determine csPCa (GGG≥2) and ciPCa (GGG=1). 5 experienced radiologists, with over 5 years of experience in prostate imaging, delineated PCa ROIs on bi-parametric MRI (bpMRI including T2 weighted (T2W) and diffusion weighted (DWI) sequences) within the training set (n₁=160 lesions) using targeted biopsy locations. Patches were extracted using the ROIs which were then used to train individual CNNs (N₁-N₅) using the SqueezeNet architecture. The average volume for readerdelineated ROIs used for training varied greatly, ranging between 1106 and 2107 mm across all readers. The resulting networks showed no significant difference in classification performance (AUC= 0.82 ± 0.02) indicating that they were relatively robust to inter-reader variations in ROI. These models were evaluated on independent test sets (n₂=85 lesions, n₃=29 lesions) where ROIs were obtained by co-registration of MRI with post-surgical pathology, unaffected by inter-reader variations in ROIs. Network performance across D₂ and D₃ was 0.80±0.02 and 0.62 ± 0.03, respectively. The CNN predictions were moderately consistent, with ICC(2,1) scores across D₂ and D₃ being 0.74 and 0.54, respectively. Higher agreement in ROI overlap produced higher correlation in predictions on external test sets (R = 0.89, p < 0.05). Furthermore, higher average ROI volume produced greater AUC scores on D₃, indicating that comprehensive ROIs may provide more features for DL networks to use in classification. Inter-reader variations in ROIs on MRI may influence the reliability and generalizability of CNNs trained for PCa risk stratification.

Proceedings Article | 16 March 2020 Paper

Radiomic features derived from periprostatic fat on pre-surgical T2w MRI predict extraprostatic extension of prostate cancer identified on post-surgical pathology: preliminary results

Rakesh Shiradkar, Ruyuan Zuo, Amr Mahran, Lee Ponsky, Sree Harsha Tirumani, Anant Madabhushi

Proceedings Volume 11314, 113143G (2020) https://doi.org/10.1117/12.2551248

KEYWORDS: Magnetic resonance imaging, Prostate cancer, Prostate, Machine learning, Feature extraction, Principal component analysis, Imaging systems, Statistical analysis, Cancer, Pathology

Read Abstract +

Proceedings Article | 16 March 2020 Presentation + Paper

A combination of intra- and peri-tumoral deep features from prostate bi-parametric MRI can distinguish clinically significant and insignificant prostate cancer

Amogh Hiremath, Rakesh Shiradkar, Nathaniel Braman, Prateek Prasanna, Art Rastinehad, Andrei Purysko, Anant Madabhushi

Proceedings Volume 11314, 113140M (2020) https://doi.org/10.1117/12.2550963

KEYWORDS: Magnetic resonance imaging, Prostate cancer, Feature extraction, Convolutional neural networks, Prostate, Diffusion weighted imaging

Read Abstract +

Proceedings Article | 13 March 2019 Paper

Radiomic features derived from pre-operative multi-parametric MRI of prostate cancer are associated with Decipher risk score

Lin Li, Rakesh Shiradkar, Ahmad Algohary, Patrick Leo, Cristina Magi-Galluzzi, Eric Klein, Andrei Purysko, Anant Madabhushi

Proceedings Volume 10950, 109503Y (2019) https://doi.org/10.1117/12.2512606

KEYWORDS: Feature extraction, Prostate cancer, Cancer, Magnetic resonance imaging, Pathology, Prostate, Tissues, Diffusion, Image registration

Read Abstract +

Decipher, a genomic test, is used to predict the likelihood of metastasis and prostate cancer (PCa) specific mortality based on expression patterns of 22 RNA markers from radical prostatectomy (RP) specimens. It has been shown to be strongly correlated with metastasis-free prognosis and has been integrated with the National Comprehensive Cancer Network (NCCN) guidelines. However, Decipher is expensive and tissue destructive. Radiomic features refer to the high-throughput computational texture or shape features extracted from radiographic scans. Radiomic features derived from multi-parametric magnetic resonance imaging (mpMRI) of prostate cancer have been shown to be associated with clinically significant PCa. In this study, we sought to evaluate whether radiomic features derived from T2-weighted MRI (T2WI) and apparent diffusion coefficient (ADC) maps of the prostate could distinguish different Decipher risk groups (low, intermediate and high). We also explored correlations between Decipher risk associated radiomic features and features relating to gland morphology on corresponding digitized surgical specimens. A retrospectively acquired, de-identified cohort of 70 PCa patients (N = 74 lesions) who underwent 3T mpMRI prior to RP and Decipher tests after RP were used in this study. The Decipher risk score, ranging from 0 to 1, was used to categorize patients into low/intermediate (D₁) and high (D₂) risk groups. A multivariate logistic regression model was trained (N = 37 lesions) using radiomic features selected via elastic-net regularization to predict the Decipher risk groups. The model was evaluated on a hold-out test set (N = 37 lesions) and resulted in an area under the receiver operating characteristic curve (AUC) = 0:80. Our model outperformed the prediction using PIRADS v2 (AUC = 0:67), but showed comparable performance with Gleason Grade Group (GGG) (AUC = 0:80). We observed that the best discriminating radiomic features were correlated with gland morphology and gland packing on corresponding histopathology (R = 0.43, p < 0.05).

Conference Committee Involvement (1)

Image Processing

17 February 2025 | San Diego, California, United States

Keywords/Phrases

Search In:

Publication Years