A frequent cancer in male community is Prostate cancer. If it is identified in early stages, then it will be curable. This cancer is diffusing all over the world including France, USA, Swedon and Ireland etc. More than 25,400 males are affected by this cancer. This gland looks like walnut. Most of the times it grows slowly in many men, unfortunately it grows exponentially in some people. It creates blood during urination and in semen. Early-stage identification needs close analysis and complete diagnosis with medications. For this purpose, many deep learning methods are suggested. In this paper, convolution layer based deep learning model has been used. Out of this, Visual Geometry Group-16 (VGG-16) model yields accuracy of 97.74% and mobile net model gives accuracy of 86.24%. This work suggests that all the cancers can be treated with the kit of deep learning models assisted software.
Top1. Introduction
Prostate cancer occurs in men's reproductive system. As per the statistics in (World cancer research,2022). All over the world, 1,414,259 men are affected by this prostate cancer. Every year 1.41 million cases and 34,700 deaths are newly reported. Out of all countries, Sweden is affected more 10,949 numbers per year. As usual with other cancer types, Prostate cancer starts with cell mutation (i.e.) when there is a small change in DNA sequence this will be extended throughout the gland. This change in gland has been taken place either slow or very fast. Some of the common symptoms of prostate cancer can be spotted in urine such as (i) trouble in urinating, serum or blood in urine, additionally some sufferings in the pelvic area, and bone pain. Anyway, these symptoms can also be seen due to some other issues. Prostate cancer affected organ is shown in Figure 1.
Proper physical examinations, which include collecting a tiny sample of tissue from the prostate gland for laboratory testing, are often used to diagnose prostate cancer. to diagnose, MRI (magnetic resonance imaging) pictures can provide detailed views of the prostate gland. Image based classifications for prostate cancer detection methods are suggested. MRI can also be used to guide a prostate biopsy, which includes extracting tiny tissue samples for laboratory testing. This approach, also known as MRI-guided biopsy or fusion biopsy, has the possibility to improve the correctness of prostate cancer diagnosis. Some screening tests digital rectal exams and pro-state specific antigen blood tests If that levels are high means there is a high probability of cancer. Levels also rise for benign conditions) are suggested. Many of the researchers try to predict the pro state cancer in early stages.
Figure 1. Difference between normal and cancer affected organ (Medicover hospitals)
For predicting from images deep learning techniques can be applied in medical field. Deep learning networks are classified as Artificial Neural Networks (ANN), Convolutional neural networks (CNNs), Generally, the process in machine learning based medical analysis generally includes the following steps:
Classification: This process involves assigning of class or label to the specific pixels or tumor portions. Segmentation algorithms can be used for this classification.
Algorithm: The set of procedures followed for correct classifications, and to predict the given tissues for the future modulation towards the disease; Region of interest (ROI) marking with correct boundaries also the task of good algorithm. Hence the functions include diagnosis of disease (tumor/ cancer), type of cancer with correct stage, and the areas where it spreads, and the possibility of spreading etc.,.
Training phase: The stage during which the specimen of image or data are given with correct label or classification. Certain percentage of data are given for training phase, this process is continued until correct classification occurs.
Testing phase: This is factual examples of every disease in medical world. Precision and accuracy of this unnoticed information decides the level of algorithm. For doing these activities in the proposed approach, CNN architecture has been used.
Table 1. Machine learning techniques and their accuracies
| Ref | Data set used | Technique used | Classifier | Accuracy (%) |
| Hassan et al.,, 2022 | Prostate-MRI-US-Biopsy | VGG16 | Gradient boosting | 87.10 |
| Nearest Neighbor | 86.90 |
| Random Forest | 87.5 |
| Support Vector Machine | 87.15 |
|
Shao et al., 2021
| PCA-GGDATA,
PROSTATEX | PCa-GGNet-v2 | Multi-class classification | 80 |
|
Abbasi et al, 2020
| Prostate MR (2018) Image database. | Deep learning convolutional neural network | SVM Gaussian+Texture+Morphological EFD | 99.71 |
| SVM gaussian+texture+SIFT | 98.83 |
| Ogino et al.,2021 | Cancer Imaging
Archive | Feature space transfer mode | Three-class classification
Two-class classification | 82.5
88.0 |
| Qian et al.,2017 | ProCDet. | Self-supervised learning | - | 91.82 |
| Liu et al.,2017 | DWI database | Deep learning & convolutional neural network | - | 78.15 |
| Duran et al., 2020 | WSI image | Convolutional neural network | | 99.98% |