Predicting Prostate Cancer Upgrading of Biopsy Gleason Grade Group at Radical Prostatectomy Using Machine Learning-Assisted Decision-Support Models

Machine learning discrimination of Gleason scores below GG3 and above GG4 for HSPC patients diagnosis

This study aims to develop machine learning (ML)-assisted models for analyzing datasets related to Gleason scores in prostate cancer, conducting statistical analyses on the datasets, and identifying meaningful features. We retrospectively collected data from 717 hormone-sensitive prostate cancer (HSPC) patients at Yunnan Cancer Hospital. Of these, data from 526 patients were used for modeling. Seven auxiliary models were established using Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Extreme gradient boosting tree (XGBoost), Adaptive Boosting (Adaboost), and artificial neural network (ANN) based on 21 clinical biochemical indicators and features. Evaluation metrics included accuracy (ACC), precision (PRE), specificity (SPE), sensitivity (SEN) or regression rate(Recall), and f1 score. Evaluation metrics for the models primarily included ACC, PRE, SPE, SEN or Recall, f1 score, and area under the curve(AUC). Evaluation metrics were visualized using confusion matrices and ROC curves. Among the ensemble learning methods, RF, XGBoost, and Adaboost performed the best. RF achieved a training dataset score of 0.769 (95% CI: 0.759-0.835) and a testing dataset score of 0.755 (95% CI: 0.660-0.760) (AUC: 0.786, 95%CI: 0.722-0.803), while XGBoost achieved a training dataset score of 0.755 (95% CI: 95%CI: 0.711-0.809) and a testing dataset score of 0.745 (95% CI: 0.660-0.764) (AUC: 0.777, 95% CI: 0.726-0.798). Adaboost scored 0.789 on the training dataset (95% CI: 0.782-0.857) and 0.774 on the testing dataset (95% CI: 0.651-0.774) (AUC: 0.799, 95% CI: 0.703-0.802). In terms of feature importance (FI) in ensemble learning, Bone metastases at first visit, prostatic volume, age, and T1-T2 have significant proportions in RF's FI. fPSA, TPSA, and tumor burden have significant proportions in Adaboost's FI, while f/TPSA, LDH, and testosterone have the highest proportions in XGBoost. Our findings indicate that ensemble learning methods demonstrate good performance in classifying HSPC patient data, with TNM staging and fPSA being important classification indicators. These discoveries provide valuable references for distinguishing different Gleason scores, facilitating more accurate patient assessments and personalized treatment plans.

PSA doubling time 4.65 months as an optimal cut-off of Japanese nonmetastatic castration-resistant prostate cancer

A multicenter study of nonmetastatic castration-resistant prostate cancer (nmCRPC) was conducted to identify the optimal cut-off value of prostate-specific antigen (PSA) doubling time (PSADT) that correlated with the prognosis in Japanese nmCRPC. Of the 515 patients diagnosed and treated for nmCRPC at 25 participating Japanese Urological Oncology Group centers, 450 patients with complete clinical information were included. The prognostic values of clinical factors were evaluated with respect to prostate specific antigen progression-free (PFS), cancer-specific survival (CSS), and overall survival (OS). The optimal cutoff value of PSADT was identified using survival tree analysis by Python. The Median PSA and PSADT at diagnosis of nmCRPC were 3.3 ng/ml, and 5.2 months, respectively. Patients treated with novel hormonal therapy (NHT) showed significantly longer PFS (HR: hazard ratio 0.38, p < 0.0001) and PFS2 (HR 0.45, p < 0.0001) than those treated with vintage nonsteroidal antiandrogen agent (Vintage). The survival tree identified 4.65 months as the most prognostic PSADT cutoff point. Among the clinical and pathological factors PSADT of < 4.65 months remained an independent prognostic factor for OS (HR 2.96, p = 0.0003) and CSS (HR 3.66, p < 0.0001). Current data represented optimal cut-off of PSADT 4.65 months for a Japanese nmCRPC.

The potential of machine learning models to identify malnutrition diagnosed by GLIM combined with NRS-2002 in colorectal cancer patients without weight loss information

BACKGROUND & AIMS

The key step of the Global Leadership Initiative on Malnutrition (GLIM) is nutritional risk screening, while the most appropriate screening tool for colorectal cancer (CRC) patients is yet unknown. The GLIM diagnosis relies on weight loss information, and bias or even failure to recall patients' historical weight can cause misestimates of malnutrition. We aimed to compare the suitability of several screening tools in GLIM diagnosis, and establish machine learning (ML) models to predict malnutrition in CRC patients without weight loss information.

METHODS

This multicenter cohort study enrolled 4487 CRC patients. The capability of GLIM diagnoses combined with four screening tools in predicting survival probability was compared by Kaplan-Meier curves, and the most accurate one was selected as the malnutrition reference standard. Participants were randomly assigned to a training cohort (n = 3365) and a validation cohort (n = 1122). Several ML approaches were adopted to establish models for predicting malnutrition without weight loss data. We estimated feature importance and reserved the top 30% of variables for retraining simplified models. The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity were calculated to assess and compare model performance.

RESULTS

NRS-2002 was the most suitable screening tool for GLIM diagnosis in CRC patients, with the highest hazard ratio (1.59; 95% CI, 1.43-1.77). A total of 2076 (46.3%) patients were malnourished diagnosed by GLIM combined with NRS-2002. The simplified random forest (RF) model outperformed other models with an AUC of 0.830 (95% CI, 0.805-0.854), and accuracy, sensitivity and specificity were 0.775, 0.835 and 0.742, respectively. We deployed an online application based on the simplified RF model to accurately estimate malnutrition probability in CRC patients without weight loss information (https://zzuwtt1998.shinyapps.io/dynnomapp/).

CONCLUSIONS

Nutrition Risk Screening 2002 was the optimal initial nutritional risk screening tool in the GLIM process. The RF model outperformed other models, and an online prediction tool was developed to properly identify patients at high risk of malnutrition.

Machine learning prediction of Gleason grade group upgrade between in-bore biopsy and radical prostatectomy pathology

This study aimed to enhance the accuracy of Gleason grade group (GG) upgrade prediction in prostate cancer (PCa) patients who underwent MRI-guided in-bore biopsy (MRGB) and radical prostatectomy (RP) through a combined analysis of prebiopsy and MRGB clinical data. A retrospective analysis of 95 patients with prostate cancer diagnosed by MRGB was conducted where all patients had undergone RP. Among the patients, 64.2% had consistent GG results between in-bore biopsies and RP, whereas 28.4% had upgraded and 7.4% had downgraded results. GG1 biopsy results, lower biopsy core count, and fewer positive cores were correlated with upgrades in the entire patient group. In patients with GG > 1 , larger tumor sizes and fewer biopsy cores were associated with upgrades. By integrating MRGB data with prebiopsy clinical data, machine learning (ML) models achieved 85.6% accuracy in predicting upgrades, surpassing the 64.2% baseline from MRGB alone. ML analysis also highlighted the value of the minimum apparent diffusion coefficient ( ADC min ) for GG > 1 patients. Incorporation of MRGB results with tumor size, ADC min value, number of biopsy cores, positive core count, and Gleason grade can be useful to predict GG upgrade at final pathology and guide patient selection for active surveillance.

Efficient 18F-fluorodeoxyglucose positron emission tomography/computed tomography-based machine learning model for predicting epidermal growth factor receptor mutations in non-small cell lung cancer

BACKGROUND

Beyond the human eye's limitations, radiomics provides more information that can be used for diagnosis. We develop a personalized and efficient model based on 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography/computed tomography (PET/CT) to predict epidermal growth factor receptor (EGFR) mutations to help identify which non-small cell cancer (NSCLC) patients are candidates for EGFR-tyrosine kinase inhibitors (TKIs) therapy.

METHODS

We retrospectively included 100 patients with NSCLC and randomized them according to 70 patients in the training group and 30 patients in the validation group. The least absolute shrinkage and selection operator logistic regression (LLR) algorithm and support vector machine (SVM) classifier were used to build the models and predict whether EGFR is mutated or not. The predictive efficacy of the LLR algorithm-based model and the SVM classifier-based model was evaluated by plotting the receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC).

RESULTS

The AUC, sensitivity and specificity of our radiomics model by LLR algorithm were 0.792, 0.967, and 0.600 for the training group and 0.643, 1.00, and 0.378 for the validation group, respectively, in predicting EGFR mutations. The AUC was 0.838 for the training group and 0.696 for the validation group after combining radiomics features with clinical features. The prediction results based on the SVM classifier showed that the validation group had the best performance when based on radial kernel function with AUC, sensitivity, and specificity of 0.741, 0.667, and 0.825, respectively.

CONCLUSIONS

Radiomics models based on 18F-FDG PET/CT modeled with different machine learning algorithms can improve the predictive efficacy of the models. Models that combine clinical features are more clinically valuable.

Intravoxel incoherent motion predicts positive surgical margins and Gleason score upgrading after radical prostatectomy for prostate cancer

BACKGROUND

Whether Intravoxel incoherent motion (IVIM) can be used as a predictive tool of positive surgical margins (PSMs) and Gleason score (GS) upgrading in prostate cancer (PCa) patients after radical prostatectomy (RP) still remains unclear. The aim of this study is to explore the ability of IVIM and clinical characteristics to predict PSMs and GS upgrading.

METHODS

A total of 106 PCa patients after RP who underwent pelvic mpMRI (multiparametric Magnetic Resonance Imaging) between January 2016 and December 2021 and met the requirements were retrospectively included in our study. IVIM parameters were obtained using GE Functool post-processing software. Logistic regression models were fitted to confirm the predictive risk factor of PSMs and GS upgrading. The area under the curve and fourfold contingency table were used to evaluate the diagnostic efficacy of IVIM and clinical parameters.

RESULTS

Multivariate logistic regression analyses revealed that percent of positive cores, apparent diffusion coefficient and molecular diffusion coefficient (D) were independent predictors of PSMs (Odds Ratio (OR) were 6.07, 3.62 and 3.16, respectively), Biopsy GS and pseudodiffusion coefficient (D*) were independent predictors of GS upgrading (OR were 0.563 and 7.15, respectively). The fourfold contingency table suggested that combined diagnosis increased the ability of predicting PSMs but had no advantage in predicting GS upgrading except the sensitivity from 57.14 to 91.43%.

CONCLUSIONS

IVIM showed good performance in predicting PSMs and GS upgrading. Combining IVIM and clinical factors enhanced the performance of predicting PSMs, which may contribute to clinical diagnosis and treatment.

Machine-learning predicts time-series prognosis factors in metastatic prostate cancer patients treated with androgen deprivation therapy

Machine learning technology is expected to support diagnosis and prognosis prediction in medicine. We used machine learning to construct a new prognostic prediction model for prostate cancer patients based on longitudinal data obtained from age at diagnosis, peripheral blood and urine tests of 340 prostate cancer patients. Random survival forest (RSF) and survival tree were used for machine learning. In the time-series prognostic prediction model for metastatic prostate cancer patients, the RSF model showed better prediction accuracy than the conventional Cox proportional hazards model for almost all time periods of progression-free survival (PFS), overall survival (OS) and cancer-specific survival (CSS). Based on the RSF model, we created a clinically applicable prognostic prediction model using survival trees for OS and CSS by combining the values of lactate dehydrogenase (LDH) before starting treatment and alkaline phosphatase (ALP) at 120 days after treatment. Machine learning provides useful information for predicting the prognosis of metastatic prostate cancer prior to treatment intervention by considering the nonlinear and combined impacts of multiple features. The addition of data after the start of treatment would allow for more precise prognostic risk assessment of patients and would be beneficial for subsequent treatment selection.

Discriminatory Gleason grade group signatures of prostate cancer: An application of machine learning methods

One of the most precise methods to detect prostate cancer is by evaluation of a stained biopsy by a pathologist under a microscope. Regions of the tissue are assessed and graded according to the observed histological pattern. However, this is not only laborious, but also relies on the experience of the pathologist and tends to suffer from the lack of reproducibility of biopsy outcomes across pathologists. As a result, computational approaches are being sought and machine learning has been gaining momentum in the prediction of the Gleason grade group. To date, machine learning literature has addressed this problem by using features from magnetic resonance imaging images, whole slide images, tissue microarrays, gene expression data, and clinical features. However, there is a gap with regards to predicting the Gleason grade group using DNA sequences as the only input source to the machine learning models. In this work, using whole genome sequence data from South African prostate cancer patients, an application of machine learning and biological experiments were combined to understand the challenges that are associated with the prediction of the Gleason grade group. A series of machine learning binary classifiers (XGBoost, LSTM, GRU, LR, RF) were created only relying on DNA sequences input features. All the models were not able to adequately discriminate between the DNA sequences of the studied Gleason grade groups (Gleason grade group 1 and 5). However, the models were further evaluated in the prediction of tumor DNA sequences from matched-normal DNA sequences, given DNA sequences as the only input source. In this new problem, the models performed acceptably better than before with the XGBoost model achieving the highest accuracy of 74 ± 01, F1 score of 79 ± 01, recall of 99 ± 0.0, and precision of 66 ± 0.1.

Machine Learning-Based Prediction of Pathological Upgrade From Combined Transperineal Systematic and MRI-Targeted Prostate Biopsy to Final Pathology: A Multicenter Retrospective Study

Objective

This study aimed to evaluate the pathological concordance from combined systematic and MRI-targeted prostate biopsy to final pathology and to verify the effectiveness of a machine learning-based model with targeted biopsy (TB) features in predicting pathological upgrade.

Materials and Methods

All patients in this study underwent prostate multiparametric MRI (mpMRI), transperineal systematic plus transperineal targeted prostate biopsy under local anesthesia, and robot-assisted laparoscopic radical prostatectomy (RARP) for prostate cancer (PCa) sequentially from October 2016 to February 2020 in two referral centers. For cores with cancer, grade group (GG) and Gleason score were determined by using the 2014 International Society of Urological Pathology (ISUP) guidelines. Four supervised machine learning methods were employed, including two base classifiers and two ensemble learning-based classifiers. In all classifiers, the training set was 395 of 565 (70%) patients, and the test set was the remaining 170 patients. The prediction performance of each model was evaluated by area under the receiver operating characteristic curve (AUC). The Gini index was used to evaluate the importance of all features and to figure out the most contributed features. A nomogram was established to visually predict the risk of upgrading. Predicted probability was a prevalence rate calculated by a proposed nomogram.

Results

A total of 515 patients were included in our cohort. The combined biopsy had a better concordance of postoperative histopathology than a systematic biopsy (SB) only (48.15% vs. 40.19%, p = 0.012). The combined biopsy could significantly reduce the upgrading rate of postoperative pathology, in comparison to SB only (23.30% vs. 39.61%, p < 0.0001) or TB only (23.30% vs. 40.19%, p < 0.0001). The most common pathological upgrade occurred in ISUP GG1 and GG2, accounting for 53.28% and 20.42%, respectively. All machine learning methods had satisfactory predictive efficacy. The overall accuracy was 0.703, 0.768, 0.794, and 0.761 for logistic regression, random forest, eXtreme Gradient Boosting, and support vector machine, respectively. TB-related features were among the most contributed features of a prediction model for upgrade prediction.

Conclusion

The combined effect of SB plus TB led to a better pathological concordance rate and less upgrading from biopsy to RP. Machine learning models with features of TB to predict PCa GG upgrading have a satisfactory predictive efficacy.