Diagnosing Malaria Patients with Plasmodium falciparum and vivax Using Deep Learning for Thick Smear Images

Abstract

We propose a new framework, PlasmodiumVF-Net, to analyze thick smear microscopy images for a malaria diagnosis on both image and patient-level. Our framework detects whether a patient is infected, and in case of a malarial infection, reports whether the patient is infected by Plasmodium falciparum or Plasmodium vivax. PlasmodiumVF-Net first detects candidates for Plasmodium parasites using a Mask Regional-Convolutional Neural Network (Mask R-CNN), filters out false positives using a ResNet50 classifier, and then follows a new approach to recognize parasite species based on a score obtained from the number of detected patches and their aggregated probabilities for all of the patient images. Reporting a patient-level decision is highly challenging, and therefore reported less often in the literature, due to the small size of detected parasites, the similarity to staining artifacts, the similarity of species in different development stages, and illumination or color variations on patient-level. We use a manually annotated dataset consisting of 350 patients, with about 6000 images, which we make publicly available together with this manuscript. Our framework achieves an overall accuracy above 90% on image and patient-level.

Tile-Based Microscopic Image Processing For Malaria Screening Using Deep Learning Approach

“…In addition, the selected model was evaluated on 1141 uninfected images collected from 50 uninfected patients [19] which have not been used during model training. This dataset is a high-resolution thick smear microscopic image and has a resolution of 4032 x 3024 pixels.…”

“…Similarly, microscopic images captured through the eyepiece of a microscope by attaching a digital camera or smartphone camera have high resolution [7], [27]. Previous works [19], [28] for malaria parasite detection leverage such HR images directly to SOTA deep learningbased object detectors. However, directly feeding HR microscopic images to SOTA deep learning algorithms to detect very tiny objects -for example, P. falciparum, is almost impossible.…”

“…Another study by [19] proposed a multi pipeline approach by applying Mask-RCNN as pre-candidate P. falciparum and P. vivax species detector followed by classifier head to filter out false positives. They evaluate their detection system using image-level and patient-level by using experimentally defined threshold scores.…”

“…The third external dataset is used to test performance of the proposed detection model on negative images. This dataset consists of 1141 thick smear microscopic images obtained from 50 uninfected patients and it was released by previous research work [19]. All the datasets are captured by attaching a mobile phone camera to the eyepiece of microscope and they differ in their staining style and imaging characteristics.…”

“…Recently, the advancement of deep learning algorithms has celebrated a lot of achievements in its application in a range of medical imaging tasks including medical image segmentation and reconstruction [14], [15], classification tasks [16], [17], and object detection [18]. Furthermore, recent studies show deep learning-based algorithms outperformed conventional image processing and machine learning techniques for detection and identification of malaria parasites using microscopic images of thin and thick blood smear [19], [20].…”